Okay this doesn't actually relate to any SM3.0 games, but I do have an observation I'd like to relate regarding SM3.0 performance.
I was fiddling around with the new (unofficial) beta FW 82.65 and decided I didn't like the driver, so I restored my system back to my previous driver which was 79.11.
I also installed the DirectX update.
The performance of 79.11 under the DirectX update is much faster than 79.11 before the update was applied (I back my system up using Drive Image, so I can "go back in time" and test).
To make sure 82.65 wasn't causing the speedup, I restored the old 79.11 backup, added the DX update, uninstalled the driver and reinstalled cleanly.
Nalu & Mad Mod Mike now run a fair bit faster and MUCH smoother than before. Both are SM3.0 demos that utilize dynamic branching.
Makes you wonder if ATi's "superior branching" is architecture related or API related... I seem to recall that one of the highups in Microsofts DirectX team left Microsoft to join ATi not long before R300's launch...