Update: Everyone effected by the problem, try the new build.
The new build works well, thanks! What was the issue?
Side note: I can now use a worksize of either 128 or 256 on OS X 10.7.4 and notice a small increase in hashrate over -w 64. May be worth mentioning in the OP instructions. I assume it's because of (one of?) the OpenCL updates Apple has pushed out.
FINALLY THEY FIXED IT! HOLY CRAP APPLE, IT ONLY TOOK TWO YEARS!
Also, the issue on AMD was AMD interprets the specification differently than I do. It costs basically nothing to fix it, and on GCN its measurably nothing.