I was more thinking that the compatibility layer would maybe suck in a whole python 2 parser, and potentially even the interpreter, operating on potentially common objects in memory. But just how different are the bytecodes for 2 vs. 3? Even if they were very different, I guess my question is, how hard would it be to transpile 2 bytecode into 3 bytecode. I'd think it would be loads simpler than what Jython does on the JVM for what I can only imagine must be a radically different virtual machine and memory model.

I guess the real answer is probably more political rather than technical.