Also remember most of the work on Rust at the moment is focused on stabilizing the language — performance is only really required to be acceptable for now, provided it is possible to improve.
Yes, there are several places where, for example, we generate sub-optimal LLVM IR, but time is better spent getting the interfaces and language spec correct for 1.0 than squeezing out performance.
I should mention that that's different than 'who cares about performance, let's toss this in.' Performance aspects are absolutely taken into account when changing the language. But that's different than the implementation itself.
