> But if you're optimising for that state, then you don't care about inefficiencies of one model or another.

Sure you do, there is noticable memory and ctx switching overhead(for large no. of them ofc) in stackful coroutines.