:man-facepalming:scala

This discussion is mostly about what the benchmarks are actually measuring and how they interact with internal optimizations performed by each effect system implementation. My opinion is there's no good metric against which a framework should be optimized, because it is highly dependent on the user application. The application should be profiled and optimized as a whole, and the user should have full control over how threads are used by the framework. As a result, missionary execution model is simpler and easier to reason about (no hidden thread pool).

At the end of the day framework performance is a minor issue compared to correctness, composition and semantics.