Hi Clare,
The warnings have been added to 3drefit's deoblique option because all it does is replace the oblique information with the closest cardinal transformation. If you truly want to 'deoblique' you need to use 3dWarp.
I don't know what to recommend for a 'best approach' because I do not know what you're doing with the data when you move to FSL. You say that you've checked the data thoroughly, so assuming that the registration in FSL is working then you're OK.
However, when epi and anatomy are significantly out of alignment because of differing scanning planes, the odds of the alignment not converging are significant. So to be careful, I would recommend that you deoblique the EPI before going to FSL.
The only downside here, is that you are adding an interpolation step on the EPI data processing. But if you're blurring the EPI later on then I would not worry much about that extra blur.
Lastly, you should verify the alignment closely, regardless of what tool you're using, because alignments are not always that good.
cheers,
ziad