3dQwarp is tricky to use on people with "abnormal" brains -- people with resections, tumors, significant dysplasia, .... In some cases, probably nothing useful can be done -- see the image in this Wikipedia article, for example: [en.wikipedia.org]
In other cases, the "-emask" option might make it possible to get a good alignment using the "normal" part of the brain. Infant brains, with lesser GM/WM contrast, might also pose difficulties.

If you have partial brain coverage, then nonlinear warping is unlikely to work well when trying to match to a full brain base, without some special efforts and experimentation.

I'm not sure what other criteria for evaluating the alignment I could recommend, other than visual inspection. I suppose one could use 3dSeg and see how the GM overlaps between subjects -- but I've not tried that (yet). It is important to inspect alignment visually, since weird things can happen when warping a 3D volume -- especially if the skull stripping wasn't good (this is why the Human Connectome Project warps with the skull on, but I've not tried that with 3dQwarp).

It is important to use an appropriate base volume. The standard MNI 152 brain average template is too blurry to be a good nonlinear warp base. The MNI 152 nonlinearly warped average (we call it MNI152_T1_2009c in our distribution) is much better for this purpose.

The default level of warp refinement should give reasonable results. It is possible to run to higher levels of refinement, but the results will look "funny". When you run to the finest level of warping on a large collection of brains, then average them, you get something very nice looking, like the MNI152_T1_2009c template. But if you look at the individual volumes at such a high level of warping, they will look "funny" in detail -- lots of strange distortions at the finest scale. The "-pblur" option was introduced to try to reduce these (among other reasons), but was only partially successful. So the level of warping refinement that you use depends on your application -- if you are making a template, the higher the better. If you want the individual brains to be "reasonable looking", then don't run "-minpatch" below 19 or so (the default is 25).

I hope this helps.