Sorry for dropping this a bit, but it seems like you are getting through it well. This sort of operation has been on the "todo" list for many years now, and we just haven't gotten to it yet. But indeed, it would mean computing an extra NL warp between the already affine aligned anat and epi vr_base datasets, and then applying it via 3dNwarpApply, along with all other xforms.
The plan is of course to allow concatenation with even prior distortion correction and NL template alignment. There are many alignment operations already, so it has to fit into the list. But still, it is just a matter of getting it done. I am sorry that we have been slow with this, but consider it a good "character building" exercise. :)
Anyway, great job going through these details. And feel free to ping us once in a while, to help keep it near the top of "the list".
Thanks,
- rick