It makes perfect sense: Rust compilers will never beat a human at scheduling every single opcode perfectly based on the deepest microarchitectural analysis short of decapping the chip and breaking out the ol' electron microscope. Whether it's worthwhile to be that efficient over a whole program, as opposed to a preternaturally tight compute kernel, is definitely questionable.
Yes it does, that is why most serious games used to be written in Assembly, until BASIC, Pascal, C and C++ compilers got enough language extensions and compiler knobs to not having to do that all the time, and yet some Assembly is still required.
