align_epi_anat.py tries to figure out whether slice timing needs to be applied to the dataset by looking at the dataset header. The easy way around this is to just run 3dTshift on the original EPI datasets and then use the time-shifted datasets as input for alignment. There is no extra interpolation that way. Slice timing correction is just a convenience in align_epi_anat.py (especially for those of us who might forget about doing that step otherwise). The other way to fix this problem is to update the header of the dataset with the slice timing information, and there are several posts about this on the messageboard, like this one:
http://afni.nimh.nih.gov/afni/community/board/read.php?1,69601,69602#msg-69602