The x/y resolution is normally given as the distance the head moves for a single microstep of the motor. You can work this out: for example, a default solidoodle 3 has (on both x and y axes):
* 18t 2.03mm pitch MXL pulleys
* 200 step/rev (1.8 degree) motor
* 1/16 microstepping

So 1 motor rev --> 18*2.03=36.54mm travel
200 * 16 = 3200 microsteps / rev
So each microstep translates to 36.54 / 3200 = 0.0114mm

So if you think about it, if you tell the machine to move to any arbitrary point, it will move to the closest step it can: which represents a quantization error of +/- 0.0114/2=0.0057mm.

This of course doesn't take into account the machine's parallelness, staightness, rigidity or backlash, which tend to outweigh this error by a fair margin.

Also note that the extrusion process itself is not madly accurate - the nozzle squishes and pulls the plastic around a fair bit, it droops and curls and does all sorts of crazy stuff, so even if you have a perfect motion control system for the nozzle, the parts would still show a level of inaccuracy due to this.

In short, to the best of my knowledge, a finely-tuned machine should give you confidence in a part accuracy in the order of +/- 0.1mm in the x/y plane (not including overall thermal contraction etc, for which you might allow about 0.5% for ABS).

Given this, there would not be a lot of value in changing the drive system for increased resolution, as the other sources of error will mask any improvement.