Effect of instruction optimization on benchmarks

From what I've heard, Intel and AMD cpus both have generic x86 instructions as well as their own "custom" instructions such as AMD 3D Now! and Intel MMX.

When a game or application is optimized for Intel or AMD processors, does that mean the game uses those custom instructions as much as possible?

In general, does the video game companies have time to optimize their code for both platforms or do they just code their game in generic x86 instructions to simply ensure general compatibility while saving time and money?

When a company has to release a game on PS3, XBox 360 and PC, I have a feeling that they might be tempted to save time and avoid the PC platform optimizations because in general the PC version already looks good compared to the console versions.

I'm curious about this because it might explain why equal priced AMD and Intel trade victories in specific benchmarks. And then I might extrapolate this and guess if my favorite upcoming games will be optimized for either platforms and that may influence my hardware purchase.

For example, the i5 750 performs much better than the 965 BE in FarCry2 which I've heard makes use of all 4 cores.

On the other hand, the 965 BE performs better than even the i7 975 in Prototype. However, I heard that the PS3 version of Prototype looks better than both the PC and Xbox 360 version.

AMD and Intel differ on SSE4.1 and SSE4a; though I don't how optimized apps react to these.