My drives were connected straight to the motherboard (using RDM) for the performance numbers I posted earlier in this thread, and I was definitely happy with those for these drives. However, I just couldn't write to the drives without errors/degrading pools (even a single-disk pool). Instead of troubleshooting that (there seems to be very little support for that method of RDM, and I don't see what I could have done wrong if the vmdk's created without errors),
I had 5 sata drives with RDMs passed in to zfsguru. Started running a benchmark and within minutes started seeing checksum errors and one drive dropping out for excessive errors. I assume something is buggy in the RDM stuff? Dropped that idea quick