As a hobby project I'm writing a VM based on VLIW opcodes, so an interpreter written in C (and in assembly for few popular architectures) could run reasonably fast without requiring JIT (i.e. instead of fetching one instruction per iteration, a big instruction containing many instructions with its respective parameters is fetched). On top of that will run a scripting language optimized for fast string processing (I'm not sure about the viability of the compiler, yet -I guess I'll have to put customized opcode combinations for most frequent load/alu/store combinations, for the specific need of the scripting language on top of that, i.e. not academia stuff-).