I'm very excited to see work like this. There's no reason to accept MapReduce as the best type signature for structuring distributed, parallelizable algorithms. Most of the justification is that it "feels right", and "just works".