Maybe distributing the fuzzing into subinterpreters can help find issues in code being tested or subinterpreters.