Will be neat if it works. The problem with parallelization is that it's hard to get performance wins. There's a lot of benchmarking and you have to be careful to stay on the right side of Ahmadel's law.

Can the analysis be automated? I hope so. That would be cool.

Haskell itself is extremely amenable to this sort of thing but no one has figured out how to implement it AFAIK. It may be possible with profile-driven optimization to at-least guarantee adding a thread doesn't do more harm than good.