I think Klasupost is on point, if one needs more optimization let it be, but they have to deal with consequences, when they consider alternatives like SIMD probably they made the call that performance needs to be better (assuming no other cause) maybe kind of inlined assembly/C/instrinct block at same source code? Same goes with the Java , I prefer it when I want to stick to discipline but there are times you need heavy computation and the trade offs forces people different architectural decisions.
*edit: also go more flexible at C code https://github.com/lxwagn/using-go-with-c-libraries