Here's the, long winded, summary:
NVIDIA GTX470's aren't good with double-precision floating point calculations (they're not needed for gaming) which means avoid MilkyWay which is mostly DP. Einstein does some DP but they've reduced it to a minimum for NVIDIA GPUs.
There's a good list here (AQUA gave up GPUs and DNETC is not available:
http://boinc.berkeley.edu/wiki/GPU_computing
Check the applications page of each project for compatibility with the CUDA driver you are using. I am using a modified Collatz application as my driver is v3.2 and doesn't work with the standard Collatz 3.1 application.
You should be able to run 2 WUs in parallel on a dual core machine and 4 WUs in parallel on a quad-core. WUs can be from different projects for example I have dual-cores with WCG and Malaria running in parallel. If the dual-cores are notebooks with fast processors above 2.2 GHz they tend to heat up a bit so you may need to reduce BOINC CPU usage to something less than 100%.... unless you like to fry eggs on the keyboard...
AQUA runs multi-processor applications in which a single WU will use all the cores you give it. The MilkyWay N-Body simulations are are also mult-processor so they will also use any available cores unless you configure them to use less than the number of cores available.