You are right - all the spatial transformations will change the acquisition time of the voxel. It's fairly tricky to follow all the spatial transformations altogether. Concatenating all the transformations and applying all of them together as we do with align_epi_anat.py and afni_proc.py would be the most correct way. Interpolation is still an important point because if you have higher resolution in voxels that cross between two acquisition times, averaging is probably not the best thing to do, but nearest neighbors aren't exactly right either. Larger voxels than the original definitely blend times too, so there's no really good solution for that. In the end, I think few if any go through the trouble of tracking times across all the processing pipeline. If it's done at all, slice timing correction is done at the beginning of most pre-processing pipelines.