Hi Chris,
If the template and the subject intensities are completely or partially reversed (CSF-dark in one and light in the other), then lpc, lpc+ZZ or nmi are usually good choices. If the sizes are very different, then the -supersize option might help too. Otherwise, I am considering adding a unifize option that might call 3dUnifize or 3dLocalUnifize that could help for some datasets. You can also invert the dataset to make a more T1-like dataset with an 3dAutomask and 3dcalc - see below. You already know these things, but for others' benefit, there are other potential problems like strange coordinates to start with or overly large FOV.
# invert T2/CT/EPI more like T1 by inverting and mask
3dAutomask -apply_prefix anat_amd.nii.gz -dilate 3 anat.nii.gz
set max = `3dBrickStat -max anat_amd.nii.gz`
3dcalc -a anat_amd.nii.gz -expr "step(a)*(${max}-a)" -prefix anat_amd_rev.nii.gz