Actually, now that I think about it some more, the real issue here is that once you have concurrency, you can just as easily have cloning happening in parallel, at which point, it doesn't actually matter how long it takes to perform a single thread-clone.

That is, as long as there is the possibility that two cloning operations in parallel threads can finish at the same time, we're toast if we're depending solely on the clock to distinguish the resulting interpreters.