As far as I understand (please correct me, djm34, if I'm wrong), the 50% gain is based on the latest opensource kernel by djm34, NOT the one on the sp_ fork, which includes about 10% improvement by sp_ and another 10% by myself.
So the gain at the end should be about 30%.
don't know for sure actually, if you can tell me what is the hashrate of the opensource code (or which sp version to use, I can do a comparison, I think sp change on blake mixing was incorporated before I gave a quote)