Hello Brianna,
For question 1, I will stick my neck out and suggest that the heptic interpolation filter has a smaller bandwidth than the Fourier counterpart, therefore making the data smoother and reducing the RMS. I could also be that Fourier interpolation might be introducing some DFT artifacts. I may be dead wrong and Bob might fire me over allegations of Fourier inadequacies. At any rate, you can/should examine the time series visually to determine if motion has been adequately suppressed and use the interpolation filter that pleases you most. (You can scroll through the time axis while gazing at the image windows, the eye is pretty good at assessing motion).
For question 2, two pass improves alignment only in the case where you have large amounts of motion between volumes such as with anatomical volumes acquired at different days. In such instances, the one pass option, which does not allow for large movements, will not converge. When you do not have large motion between volumes (like during a time series scan), the two pass does not improve convergence.
cheers,
-ziad