Give compilers some time and the optimizations will be "free" to developers.
Besides, if your application is one where performance is a serious issue, then you are going to spend some time optimizing rather than adding features. For some applications, performance is a feature.
People always seem to forget this in terms of CPU optimizations. It's both good and bad, though. If you're pegged to a given version of a compiler, the point releases likely won't see any optimization while the major releases will likely include additional optimizations for newer chipsets. And I'd wager AMD will be working to get optimizations into both open source compilers as well as things like Visual Studio. One thing to note is that switching to a newer compilker will likely be easier for most of Linux-land as opposed to Windows land. Upping the compiler version with your code compiled in GCC is often much easier than switching to a newer version of Visual Studio due to both cost and dropping older operating systems (the latest won't support Windows XP and Vista for instance, which may matter to some specific business lines and consumer areas).
> Would you do it if you had the choice between that or adding new money making features/fixing bugs?
Of course I would, because people are going to benchmark my software against my competitor's software and choose the one that runs faster on their hardware.
Especially if my software is a compiler and people are benchmarking the programs it produces against other compilers, which is where most of the optimizing actually happens these days.
Besides, if your application is one where performance is a serious issue, then you are going to spend some time optimizing rather than adding features. For some applications, performance is a feature.