Have an update. Figured I'd bump.

Reference OEM 7970, 13.3 beta, SDK that came with it. CGMiner 2.11.4, Diablo kernel, Worksize 256, Default vectors (1), intensity 11, 2 GPU Threads (default). Don't need to specify any other options aside from pool and GPU speeds. 1185 Core, 300 memory.

713.2 MHash/sec, or ~0.60186 MHash/MHz.

Running with 1 GPU thread was about 700 on the nose. Also, I have not experimented with all the combinations quite yet.

Edit: On another 7970, I was able to get slightly different core clocks, namely 1220/300, and I tested other kernel combinations and let them hash out for about 4-8 hours to get a good average that wouldn't fluctuate. Increasing vectors decreased hashrate significantly. Increasing threads beyond 2 either decreased hashrate significantly or caused a driver crash and recover. The default GPU threads (2) and default vectors (1) seem to be optimal.

730.1 MHash/sec (0.59844 MHash/MHz) with poclbm kernel and worksize=256
731.3 MHash/sec (0.59943 MHash/MHz) with poclbm kernel and worksize=128
728.1 MHash/sec (0.59680 MHash/MHz) with poclbm kernel and worksize=64 (This is the default setting.)
730.8 MHash/sec (0.59902 MHash/MHz) with diablo kernel and worksize=256
731.7 MHash/sec (0.59975 MHash/MHz) with diablo kernel and worksize=128
726.6 MHash/sec (0.59557 MHash/MHz) with diablo kernel and worksize=64

The hashrate per MHz seems to be slightly lower, probably because of a lower RAM:Core ratio. I will increase it slightly to see if I can eek out a slightly better MHash/MHz.