I looked into the T1-image quality thing too.
I personally think (look at cerebellum) that the T1_al_junk (left) looks better than the anat_final.test image (right). I want to remember those warped T1s (anat_final and anat_final_w_skull) looking more clear. We do use -volreg_align_e2a \.
Doesn't it look kind of low-res? Input T1 is > 1mm reslution isotropic and and anat_final.test is 2x2x2 mm. In our other analysis pipelines (very similar scripts) anat_final usually is of 1x1x1 resolution. Perhaps this is unrelated but better to mention it. Attching an older anat_final too, the original T1 there had 1x1x1 resolution and anat-final had 1x1x1 as well.