"It can compile itself" is an interesting benchmark of language complexity, and we now have of C and Go, so I wonder how the other common languages would compare. Of course you can "cheat" by choosing a subset that reduces the complexity of the whole.

As you can see, Mugo uses its own very inefficient ABI that’s nothing like the x86-64 ABI – the standard ABI puts the first six “cells” (64-bit values) in registers.

Considering that stacking args is the norm for 32-bit x86 and few other architectures, it's not that bad, especially since CPUs have special hardware to handle push/pop.