... and, since EPI is noisy, if you don't think there is much warping/stretching difference between the acquisitions and you are worried about fitting to noise (or, in this case, warping to noise) in the EPIs, you might also consider restricting the degrees of freedom of the affine fit. For example, you might think about using '-warp shift_rotate', so that your brain doesn't get stretched to fit noise in the skull or something.
--pt