> Does applying the polart function and computing percent signal change
> completely remove the problem of scanner drift? Or are there additional
> factors associated with scanner drift that the polart function and percent signal
> change fail to control?

Ideally modeling the low frequencies in the signal through polynomials fitting will take care of all the drifting effect and within-run scaling can remove the cross-run variability, but that's rarely the case in reality: "Essentially, all models are wrong, but some are useful" (George Box). In addition, there are other factors such as behavioral variability across runs as you've just mentioned. So if the experiment design permits, I would try to balance all the conditions and evenly distribute all the conditions among different runs. That way if something goes out of whack in the model, at least all the conditions are equally affected, and, more importantly, the resultant effect on those between-conditions contrasts would be minimal.

Gang