But wouldn't the standard optimizations be put into the pcode that gets JIT compiled later? I can't imagine that the pcode that gets sent to the end user is a line by line translation of your source.
I would certainly think so. I don't know all the workings of .Net compilation, but I know Microsoft have sunk good money into it, and performance is improving with each new release in my experience.
The thing about pre-optimising the pcode, is that it may limit the other argument about optimising for specific CPU features. If you start tailoring your pcode, then you may lose out on the ability to compile it optimally for a specific platform. In practice this is probably not much of an issue, since they only support x86 CPUs which usually have MMX or SSE.
The more CPU specific level of things, like let's say AMD has a few instructions that execute faster on their CPU than another equivalent set on an Intel, well, I don't think they would do that at all, but I may be wrong.