Hi!
The way I see it, aside from exceptional cases, separating actions into different finite states provides clarity and breakpoints but I haven't seen a loss of performance.
Unless you use a send "next frame event" action and unless I'm wrong, everything will be squeezed within the same frame and if you go too far, you'll hit the 1000 loops error thing (you'll notice frame rate dropdowns close to small hundreds of loops already anyway).
I suppose you could do a test with two dummy Fsms to compare the speed of both methods; one with all actions in one state and the other with several identical states with one copy of the same action.
I'd go with ten actions in the single state and in the second Fsm, ten states with a single idem action in each one of them, then loop both Fsms, add some counter and a display on uGUI text in both too at the end of the loop before it reboots, and run them; wait many seconds then full minutes and see which one seems to lag behind the other.