Now everyone's going to want one. And while we're dreaming, how about a hardware hypervisor with a pluggable instruction translator? Actually, if one can make a CPU expansion, why not a lightweight hypervisor? Can't context switch the custom chips, though, so I guess in the end you'd end up with a Minimig-thing capable of hosting multiple instances of the kernel. (Has anyone tried that in software with an MMU-based system?)