C/C++ to LLVM IR to asm.js to C benchmarks

It's ~3x slower than native on my test (native vs asm.js vs PNaCl): https://github.com/bkaradzic/bgfx#30hz-test-for-browsers

If the entire codebase gets through to this process, it seems like it would be no effort to port to another architecture (arm, mips, etc.)

Not saying it is an optimal way to cross compile.