I've read somewhere about TF2, and other Source Engine games, not being optimized for CPUs. Particularly AMD CPUs as they do not have the Intel architecture that is better optimized for single-threaded programs. I've got an old Phenom X4 II 955 @ 3.44 that gets between 50 - 132 w/ DX9Frames. Intel CPUs should in theory get better frames on TF2 because it was just coded that way, while other non-source games could get better with AMD.
Here is a good explanation of how CPU architecture works:
This is an hour old so my post will likely be buried but the main difference you are looking at is single thread performance vs multi-thread performance.
AMD cores special in doing several things at the same time while they are overall bad at doing each thing. They 'generally' make up. The math is simple. If 8 people do the work 70% as efficiently as 4, your still going to get more work done then the 4.
How this works is intel has better IPC (Instructions per Clock) and slightly better branch prediction, combined with a shorter pipeline that allows for quicker pipeline flushes then AMD.
What does that all mean?
Instructions per Clock generally refers to how fast a CPU can preform a instruction (or how many instruction per hz it will preform).
Branch prediction has to do with conditional jumps (if you think if statements), branch prediction attempts to predict which way an if statement will go so that the CPU can queue instructions up ahead of time to be executed. Both CPU's are very good at this, above 95% accuracy. But when they mess up you need to 'flush the pipeline'.
Pipe-lining is the 'to be executed queue' (is basically the simplest way of explaining it, generally its a queue that's continuously being partially executed (in steps) until it reaches a completed stated). Pipe lining is really only possible with branch prediction, since it allows for instructions to be queued without knowing if true or if false ahead of time (the CPU doesn't actually read or calculate the if before hand, it does funny mathy-type stuff).
Flushing is when branch prediction is WRONG, and you've accidentally queued up about a dozen things to do, that OOPS! Don't have to be done :X Now you need to push everything out of the queue at a rate of 1 item per clock cycle.
Generally speaking intel's core line is right in the 'sweet spot' of pipeline length vs. branch prediction error. Going to long of a pipeline and you'll need to flush to often and hinder performance. Go to short, and its not fully worth it (not as large of a performance gain).
So this is a round about way of saying:
What do you need?
Generally speaking an FX octo-core will serve you better if do gaming + things. Like streaming, while gaming. Or watching movies, while playing planetside2, while running mumble (because I like to party). FX cores like to do everything at once and won't really hinder your game for doing a lot of things at the same time.
Core umm cores don't as much. Quad-cores don't mind parallel tasks. But intel cores run single threaded applications a lot better. Skyrim and Starcraft for example very much prefer a Corei5 to an FX-8350 simply because their dev's were lazy [sic.] they were coded for a single thread.