That is why I am doing this now and not later. I didn't get much feedback on Lua in the beta phase. There was very little interest in it, actually. I had the choice to make between a simpler less powerful system that was a nice toy, or a more powerful system that was less forgiving. I chose the first. Then when it was released it became apparent that lots of people were very serious about using it, so I quickly decided a single state system was more appropriate.
It's not the way I would have chosen to make decisions, but as soon as it became apparent it was better to do it before anyone had committed any code. I probably do more trial and error than some other engines I have seen, but I also think we come up with better designs in the end. The Lua stuff is very new, so I would not look to it for stability right now. The main engine API has been quite stable for the last 6 months or so, and I don't see any changes happening there.
Onwards!