Actually in 3d Mark, tests are done independently which naurally does not depict the real world performance. A balance average between physics simulation score (by firing independent threads for each CPU core) and 3D rendering within one bencmark would be a good idea.
In real world 3d applications, a CPU is used for a lot of things not just Physics like caclulating net code, game data and so on, like for example MS Flight Simulator X is more CPU dependant than GPU which means the framerate factor depends on how good your CPU is. Also it depends on how well the DX 10 API mainly Geometry shader and vertex buffer code is implemented in the benchmarks because DX10 is a new architecture and so is the driver model. They should follow approved coding standards. Since we don't know that (yea i believe in uncertainity principle), I consider Vantage to be a timepass thingy

cuz i like graphics and colorful stuff.
Not only that in a few months from now, NV is going to integrate PhysX - so what will happen to CPU score then? I tell you this whole benchmark thingy is one big inaccurate data...