Looking at your example schematics I think there are many issues which should be discussed:

1. Voltage regulators always need a local supply decoupling, next to their pins. LM7805 is uncritical refering to decoupling, compared to many of the low drop regulators, nevertheless there should be a proper decoupling, giving low and resonance free impedance from lowest to highest frequencies. Especially between 100kHz and about 10MHz impedance should be low, because here ringing can occur, due to internal feedback loop circuitry of LM7805. So, I would suggest an aluminium electrolytic of 47µF/25V paralleled by 100nF/X7R at output and 47µF/25V...220µF/25V paralleled by 100nF/X7R at input of regulator. 47µF, for lower currents than 100mA and 220µF for higher currents.
The reason, that electrolytics shouldn't be much higher than above values, is that if input capacitor is in the same range as storage capacitor, relevant rectifier currents will be able run to the regulator forming a larger current loop than necessary, producing lots of aviodable interference.
At output electrolytics shouldn't be much greater than about 100µF, just because the LM7805 wasn't designed to drive such large capacitances at its output.

You might think, that using tantals for the supply decoupling is better, because impedance is much lower. That's right, but much lower impedance can cause a problem, when being paralleled by a second cap. Then, a resonance can be formed, which can result in a heavy maximum of impedance! For the frequence of this resonance, there doesn't exist proper decoupling any more!!
This phenomenon is briefly discussed at page 13 of this datasheet

http://www.analog.com/UploadedF...D797_e.pdf

The resonance is formed by the equivalent series inductance of tantal and capacitance of ceramic cap and is typically in the MHz range. Only if the bigger cap shows some relevant equivalent series resistance, than this resonance is sufficiently damped. This is the case for a standard 47µF/25V aluminium electrolytic, for which the series resistance is about 0.5Ohm, but not for a 47µF tantal, for which the resistance is about 0.1Ohm.
(Take care, aluminium electrolytics disgned for switching power supply applications can show some lower equivalent series resistance.)
As a rule of thumb, equivalent series resistance of bigger cap in such a parallel combination should suffice the following condition

R > SQRT(2 x L / C)

where L is equivalent series inductance of bigger cap and C is capacitance of smaller cap.

An example:
Through hole mounted 47µF tantal shows about L = 5nH. In combination with the smaller cap of 100nF we get R = 0.3Ohm. So, the tantal will show a resonance.
The aluminium electrolytic of 47µF/25V shows about L = 10nH. In combination with 100nF we get 0.4Ohm. But because equivalent series resistance is about 0.5Ohm, no resonance will occur.

When using tantals, then either small capacitances (<=10µF) should be used when paralleling them with smaller caps, or the smaller caps should be omitted. This is possible, if SMD tantals are used, then its inductance is small enough so that no additional cap for the high frequencies is needed.

Very bad is to parallel 1000µF by 100nF. This will cause a huge resonance! If such high caps must be paralleled by smaller caps, than use a low inductance type for the electrolytic (radial, no axial leads) and parallel it by 220...470nF.

So, 47µF/25V or 100µF/25V paralleled by 100nF/X7R can show very nice decoupling properties.

2. If an additional LC-filter at input of regulator is used, the same is true about resonances and ringing. One remedy is to insert a resistance in series to inductance or to put a resistance in parallel to it, like it's briefly discussed in this application note

http://www.web-ee.com/primers/files/u89001.pdf

Paralleling a resistor is only a proper option, if an additional series resistor would cause a too high voltage drop in combination with higher currents.
Nevertheless, I would not recommend this option, because lots of dampening is wasted by this methode. When using a series resistor, on the other hand, full impedance of inductance comes into play and very high dampenings can be achieved.

It's a good design practice to keep resonances low by a careful choose of L and C of LC-filter. As the same formula as above is valid for determining of ringing, namely

R > SQRT(2 x L / C)

it can easily be seen, that lowering L and/or highering C lowers minimum R, where here R is the series resistance added to L. So, when properly choosing L and C minimum R needed to avoid ringing can be kept very small.

Where higher currents are flowing, it's often wished to avoid this added series resistor at all, for instance at input of regulator: If big electrolytics are used enormous inrush currents can flow when turning-on the power, needing a rather strong resistor. Also, some unneccesary voltage drop might occur. Here again the relevant equivalent series resistance of aluminium electrolytic can help.

An example:
We want to avoid the series resistor and want to use 100µF at input of regulator. What inductance L should be used then?
From above formula we can see that L < 10µH is a good choice. Then mimimum R to avoid ringing is smaller 0.4Ohm. This is just about the equivalent series impedance of 100µF/25V aluminium electrolytic. So, no further series resistance is needed!

So, at input of regulator could sit a LC-filter consisting of an inductance of 1...10µF and a parallel combination of 100µF/25V and 100nF/X7R. Remember, that the caps must sit nearest to regulator.
If the storage cap is farer away from this LC-filter, then an additional 10...47µF/25V aluminium electrolytic should sit at input of LC-filter, forming a pi-filter. The reason for this measure is to not contaminate L by the unpredictable inductance of wiring to storage cap, because serielling of inductances is similar disadvantageous as paralleling of caps. Again resonances and ringing will develop, but here showing a minimum of impedance. Only the pi-filter configuration can guarantee, that dampening of filter isn't violated by complex wiring impedances.

3. LC-filtering at output of regulator is much more problematic, because here no relevant ringing and voltage drops are allowed, because it cannot be regulated out as it can for ringing at input of regulator.
In any case, the use of an LC-filter doesn't mean that decoupling caps at output of regulator can be omitted. They are still necessary!
Also, the LC-filter shouldn't sit next to the regulator, but directly at supply voltage terminals on your micro board. Otherwise the wiring between regulator and micro board becomes an antenna emitting lots of high frequency noise!

So, the LC-filter has to be mounted on your micro board. Again a pi-filter configuration should be choosen, for the same reasons as above. To achieve highest dampening the new ceramic high caps should be used, something like 10µF/10V/X5S in 0805 package. With these caps no paralleling is needed to achieve perfect performance at high frequencies. Equivalent series inductance is about 1...2nH! Equivalent series resistance is about 0.01Ohm. These ceramic high caps offer the best dampening performance ever!!
As these ceramic high caps don't show any relevant equivalent series resistance, dampening of resonances of LC-filter must be done by the help of additional series resistance.

An example:
When using L = 1µH and 10µF cap, then above formula shows, that added series resistance should be about 0.47R to avoid ringing. 100mA causes a voltage drop of only 0.047V, which is only 1% of 5V and which might be acceptable for your application. You could eliminate all eventualities, if you select a 5V regulator for your application giving at least 5.0V at maximum load.

What inductances to choose?
You could use "7427932" from Wuerth:

http://www.wuerth-elektronik.de/we_web/em...d=4&id=284

Of course, you can also use a common high current soft ferrite bead in 0603...0805 package. But 100mA can show some relevant saturation of permeability, resulting in a loss of inductance. So, the "7427932" could be the better choice.
In any case, do only use soft ferrite types. Only these show superior dampening performance in the range of 10...1000MHz.

1µH plus 10µF yields a corner frequency of about 50kHz (Thomson-formula). If you need increased dampening at much lower frequencies, then you must use a bigger inductance, about 10µH or so. SMD ferrites showing such a high inductance are seldom to find. So, a through hole 6-hole-choke might be appropriate. To achieve the same ringing suppression with 0.47Ohm resistance, filter capacitance should be 100µF. So, your pi-filter could consist of two 100µF/25V aluminium electrolytics paralleled by 100nF/X7R and a 6-hole-choke of about 10µH (estimated from low frequency performance) in series to 0.47Ohm resistor. Corner frequency will be about 5kHz, then.

You might ask now, whether the 0.47Ohm resistor can be omitted again, because of the finite equivalent series resistance of aluminium electrolytic. This is true, but additional series resistance gives better performance, means lower supply ripple. At input of regulator this is irrelevant because the regulator filters it out, but at output of regulator lowest possible ripple is needed.

Good luck,
Kai