We try, we try, dear Ant. But you don't follow our suggetions! What was wrong with the link Andy gave you?

http://focus.ti.com/lit/an/sbaa050/sbaa050.pdf

Also, I think there are many misunderstandings with your point-of-view. Noise and drift of your reference voltage is totally irrelevant, for many reasons: If you follow the link from Andy, then this drift will not have any influence on your measurement, because the constant curent source AND the ADC will undergo the same drift. So, the influence of drift of reference voltage is vanishing.
Also, as the reading of PT100 is changing only very slowly, you can reduce any relevant noise by taking the average of readings, by using the method of moving average, done in software.

Finally, if you really want to achieve a precision of two decimals with resistance measuring, then you need very very precise resistors, not showing the least long term drift and temperature drift!!

An example:
0.01 Ohm refering to 100 Ohm is 100ppm (parts per million, 1/1,000,000). As you need at least three resistors for the processing, one in the current source and two in the amplfier, then each must not show a higher drift than 60ppm, because SQRT( 60^2 + 60^2 + 60^2) = 104. If you now assume, that your ambient temperature may change by about +-10K, then you need resistors showing a temperature drift of less than +-6ppm/K, not mentioning any long term drift!!

So, wanting to achieve such a high precision is rather unrealistic. The more as we didn't take into consideration the many other error sources. I think a precision of about +-0.05 Ohm is more realistic, giving you a precision of nearly 0.1C.

Kai