I have an msi GTX560 Ti Twin Frozr-II that seems to be running slower than it should - especially when compared to my GTX570 and Tesla C2050. Wikipedia shows the 560 doing around 1.2 TFlops, the 570 at 1.4, and of course the Tesla at 1.03. That performance seems to be accurate with the 570 and Tesla, but the 560 is benching (with CUDA-Z) around 785 MFlops and does indeed take twice as long to run workunits. The core speed seems to be right (880MHz core / 1760 Shader), and I have now tried it on two different computers with the same result. Drivers are latest 275.33. Is this normal and Wikipedia's numbers full of it? Is the 570 really supposed to be TWICE as fast? Should I RMA it?

The GPUGrid client can use only the 2/3 of the GTX 560 Ti's shaders, that's why it's much slower than you expected based on the clock speed and the number of CUDA cores. I'm not sure if this is a sufficient reason for an RMA. See this message and this thread for more details.

After checking your machine with GTX570 I can see that you do not use the Swan_Sync enviroment variable. If not used, GPU load depends on CPU usage by other applications. Try to set it, free one CPU core for GPUGRID, and then compare your times to my times (GTX560 core 905 MHz, XP x64, the same driver).