Back in version 20.2.x of fMRIPrep (release notes for subsequent year’s version, 21.0.0), pepolar-style susceptibility distortion correction was done with AFNI’s 3dQwarp. From what I can tell, 3dQwarp is not sensitive to total readout time (not confident about this, so would welcome corrections). When applying the resulting warp to a bold image that has the same total readout time (as defined by FSL), that seems fine, but there are situations in which that time can differ between the fmap epis and the bold images (e.g., the fmaps have a longer echo time to maximize distortion, but then the bold has something smaller). Did older versions of fMRIPrep adjust the estimated distortion according to differences in total readout time?
Sorry to ask about the details of such an old version!