Yes, that's very odd. At least they could have used a case statement, which a C compiler will be able to optimize into an efficient jump table.

I remember reading about a VM (I remember it as being Google's V8, but that one compiles directly to machine code so probably I'm misremembering; maybe it was Strongtalk?) which basically ditched the classic bytecode loop (grab next instruction, check what it is, evaluate it, repeat) and instead implemented each bytecode type as a function and somehow munged the instruction pointer or modified the VM's machine code itself so that the CPU would move directly to the next instruction's function without jumping. Damn, I wish I could remember the idea. Anyone know what I'm talking about?