I recently gave a talk at ZuriHac on the pitfalls that one can incur on when trying to parallelize Haskell software:
Note that, as one helpful YouTube commenter notes, I wrongly talk about “expected linear speedup”, when really I mean “expected 1/n speedup”. The message of the talk stands, I was just confused when preparing the slides.
See also experience report on FP Complete’s website on this topic.