I found that there is a recommendation in another post to use the internal blur feature of 3dQwarp on both the template and input brain (-blur 3 3) when using the TT+N27 template. The current implementation in auto_warp.py (called from afni_proc.py) currently uses blur for the input but not the template (e.g. -blur 0 3). Suggesting that it would be better at performing a warp to a blurry template than a sharp template.

For what it's worth, I'm still getting very good registrations via afni_proc.py/auto_warp.py to the TT_N27 template with only a few exceptions. I assess the registration via 3dedge3 and in the AFNI viewer. I have tackling the few problematic warps on my list of things to investigate. In the meantime, if you feel enterprising, I think you could modify the line in auto_warp.py to change the default from -blur 0 3 to -blur 3 3. That may hold you over until an update to auto_warp.py enables different options.

[afni.nimh.nih.gov]