Several questions come to my mind:
1. You're seeing these bumps on the fitted model, but do you see
the same bumps in the original data?
2. You are still analyzing these as a block design, correct?
3. I don't know nuthin about 3dfim+, but with 3dDeconvolve, at
least, you don't need to resample your data, it (or 3dD + waver)
can allow for that automatically, and I think that would be preferable
to resampling. For one thing, resampling would throw off the
DOF calculations.