2) While most of the CBF to T1 registrations using the transformation for the PD to the T1 image worked reasonably, there are a few where the registration is a little off (almost exclusively requiring some translation in the SI direction) - what is the best way to correct the registration and generate a better fitting transformation matrix? (either manually or using some other tool)

What did you use to do the registration?

3) Do I not need to use seg-erode to minimize partial voluming?

You may, but then you may lose an entire ROI, esp in cortex. I'll have some PVC software for the next release.