I ran recon-all on few subjects but noticed that much of the temporal lobes were excluded from segmentation, so I added control points to missed white matter voxels in this area, and subsequently ran 

recon-all -autorecon2-cp -autorecon3. 

The control points fixed much of the segmentation errors in the temporal lobes; however, errors in segmentation of other areas of the brain -- far from where I had placed control points -- were introduced. 

Attached are images of the same subject and same coronal slice, before and after running -autorecon2-cp -autorecon3, and, as you can see, much of the right superior frontal lobe is now missed.

Is there any way to fix this, other than by simply adding more control points to the newly missed areas? One note is that the cores I am running these subjects on have a memory limit of 3.75 GB and I see that the recommended memory per subject is at least 4.0 GB. Could this be the problem? 

Also attached is the recon-all log.

Thank you!

Victor