hi,

I still disagree.
First of all, you eliminate my first case - it still indicates the reason.

Now about second one. Look at the figure below:

It is not a schematic, just an example.
As you can see, if grounds are connected together via ribbon flat cabble (RFC from now) then the current is divided in two streams: I_A and I_B. And I_B is not only that currect you said "that every signal line must have its ground return current as close as possible next to it" about. In fact, I_B is the part of all current which comes to right board from V_cc. And as many wires of RFC are used for this "loop-back" as highter I_B because we know that parallel connection decrease sum resistance.
Now just imagine that the load has not constant value, that it may be switch on/off quickly and you understand that I_B may change its value and even direction (!) at any time. And these changes will have nothing with signals which are provided via the same RFC. As result, these ground wires influence over signal ones and produce random noise and pulses which may result in malfunction.
By other words, you should remember those right words you said as well:
- make all ground connections at one and same point;
- do not make ground loops.

Yet another reason to not connect grounds at both sides:
Look at the figure again and find the point Z. Just imagine what happenes if the load is about some Ampers and the ground wire is broken at the point Z (for example, due bad contact). I think you already see in mind alot of smoke from RFC (=

Now I wish come back to your words that every signal line must have its ground return current as close as possible next to it. But why do you think that using common ground through RFC helps for it? Look at the figure again and try explain me why should the current through base-emitter of transistor come back via I_B? For me, it may do the same via I_A because both grounds are connected together inside power supply.

Regards,
Oleg