Since all of these concurrency approaches’ pain points are around shared mutable state, maybe it’s shared mutable state that’s the pain point.

Yup, this is the conclusion of Joe Armstrong and why Erlang/Elixir/BEAM just doesn't allow mutable state, other than process dictionary which is inherently non-shared.

That observation doesn't really contribute to a comparison of dedicated threads and async/await, though, because the same is true of async routines.