Discussions of technical feasibility aside, getting a sufficiently large installed base to make this interesting is (as usual) the "fun" part; your plan for get

Discussions of technical feasibility aside, getting a sufficiently large installed base to make this interesting is (as usual) the "fun" part; your plan for getting to critical mass against an available and "good enough" solution. As for previous bytecode approaches to consider, there are the Lisp Machines, or the Burroughs B5000 Algol box, or the EFI byte code (EBC) interpreter and the EBC I/O drivers, or UCSD pCode, or all the gonzo things you could do with the VAX User Writable Control Store, the JITs underneath Java and Lua and other languages, or... And HTTP is printable, which means you're working within character-encoding constraints or with escapements. There are the not-inconsequential security requirements. And then there's the question of how a provider might make money with this, if you're not undertaking this effort for free. Have at...