An easy test is to start down/over clocking your cpu and see how it affects performance in games. If its a large bottleneck you should see a decent change in performance by down/over clocking. How much of a change? I don't know, but I imagine the delta-performance per delta-cpu-clock(dp/dc) will be related to how bottlenecked the system is for a particular app. A large bottleneck *shoud* show a large dp/dc in comparison to a system with a small bottleneck, while a cpu which is not bottlenecked in a game *should* show absolutely no gains in overclocking.
You could try predict when a certain gpu becomes the main bottleneck by graphing performance vs the clock speed of the cpu. You can guess that most processors won't get above 4Ghz per core, and since you can use your cpu to create data up to 3.344 Ghz(relevant to your cpu architecture) you can try extrapolate the performance gains if it were possible to overclock your 3.344Ghz processor even faster(or replace with a faster chip of same type). I imagine this graph should show diminishing returns to just overclocking the cpu more, and at some point you decide that overclocking the cpu more doesn't make sense since you are limited by the gpu. If you find that the gpu bottleneck occurs a decent margin past your max overclock then buy a better cpu, otherwise buy a better gpu.
The bottleneck test will be very dependent on which game you use for the test, so test with whichever game you want to play with decent performance.
This is just raw performance though and doesn't represent the actual performance of the system. I imagine more cache/ram/timings(as dr emulator (madmax) mentions)/??? can help make a system have a higher minimum fps then another system even though they both have the same average fps.