I can't think of a good reason. Nevertheless, people sometimes do upsample the EPI time series to the anat resolution, producing gigantic datasets in the process. Reasons that are given (from memory)
a) They want to analyze their EPI data directly in Talairach space (in this case, they resample the EPI time series into +tlrc coords).
b) It intuitively makes sense to them (to treat the two types of data the same, I suppose).
I don't understand these reasons, but there they are. --
bob cox