I have recently compared the spatial normalization using @auto_tlrc in AFNI, the manual marker placement in AFNI, and the normalization in SPM5 on a random sample of older (65-85 y.o.) subjects T1 anatomy. It appears that the SPM5 auto normalization algorithm provides a surprisingly accurate warp in most cases (qualitatively better than @auto_tlrc in this random sample of elderly). Perhaps one difference is that SPM5 is performing a higher order warp (although I don't know for sure). Because I am new to the automated methods in AFNI, perhaps there are settings that I'm not aware of that will allow for a higher order warp that may produce more accurate results. I didn't see anything in the @auto_tlrc manual.
I also realize that a higher order warp can produce other non-optimal warping artifacts. I know nothing will put every gyrus and subcortical nucleus in the exact same place (especially in this highly variable sample of subjects) but am just trying to find some settings in AFNI's automated toolbox that will rival the SPM5 results.
By the way, the manual warping in AFNI works great on all subjects and appears to be more reliable across subjects than the SPM5 algorithm.