In general every Monad can be expressed "interpreting" the free monad. This relaxes the "one-shot" restriction, and can be implemented using delimited control. One-shot means faster performance though and is still useful for many things - and that can be implemented using coroutines.
If you acquire some understanding of what this means then you'll have a very good idea about the expressive power of what you can use coroutines (with nothing more) for, so it's very interesting.
Haskell ends up being a bad place to talk about this (or a great place, depending on your goals) because due to laziness you get to write a lot of structures which look inductive but end up being able to express coinductive structures.
From memory and intuition, if you're working in a strict language (or better yet, something like Agda where the distinction becomes very sharp) then you end up finding that continuations make for good coinductive "free" structures and the free monad (and its ilk) make for good inductive free structures.
The distinction between inductive and coinductive types is fairly subtle and hard to see in most languages where those distinctions are blurred, but broadly you can think of inductive structures as ones that are, in principle, finite and coinductive structures as being those which may be, in principle, infinite.
For example, a linked list is inductive. If you're looking at one cons cell of it you can't prove that, you may have to chase pointers for longer than your patience allows, but at least in principle there is an end. A stream is coinductive, because it instead suggests a generative process.
