I am not convinced that explicit state machines are a good way to write code. Using core.async you get an implicit state machine, while preserving the illusion of sequential code, which is much simpler from the developer point of view.
As long as you don't want to interact from said "sequential" code with the world. Then you get to deal with all the fun of asynchronous programming. Including but not limited to race conditions, data corruption and so on.
I'm not sure what you mean? Much of my core.async code deals with things like network requests.
Core.async supports Go-style channels so you are trading off speed (it has a runtime) for concurrency-correctness. It's a good fit for network requests since the bottleneck is elsewhere.