For the image that needed manual repositioning, the result you get certainly seems to be related to the fact that the image needed repositioning. As for the results from the repositioned image, I'm not sure about the cause of the rather poor result. Skull-stripping too tightly can sometimes cause problems. I don't tend to test the SPM segmentation against skull-stripped images as it was intended to be run on original images (try running most segmentation algorithms on images with skull on, and see how well they do something they were not intended to do). Low quality scans (or severe pathology) can also segment rather poorly, although they'd need to be pretty bad for that to happen.
Best regards,
-John