I believe you are correct. The transformation to Talairach, whether done from oblique or standard planes, should be equivalent. Not rotating the EPI dataset until after the individual subject results is probably a better idea. In looking at your dataset, it does seem the rotation/translation is off by about the resolution of an EPI voxel to the right and anteriorly relative to the anatomical dataset. This may be a limitation of the resolution of the original datasets, but I'll take a look at this. It's adjustable by using the Nudge Dataset plug-in. In any case, the attached JPEG file contains the images of the slightly nudged datasets over the anatomical MP-RAGE dataset.