I have an impression you’re talking about animation state machines and not hierarchical state machines for gameplay logic.
On that note, I have no idea what you’re on about in terms of optimization in either case, can you elaborate?
In both cases they’re super cheap to evaluate, no?