I'm not entirely sure how machine code(or LLVM bytecode, if you'd prefer) produced by a program written in ML is at all inferior to the same machine code(actually, probably less optimized machine code, given that ML is a high level language in which it's easier to code optimizations than in C) produced by a program written in C.

Comparing ML to Perl or Python is ridiculous. There are a number of native optimizing compilers of ML. ML is closer to C++ or Java or CL performance-wise than to those languages.