Yes, and this line says a great deal. If, by bad fortune for casinos (good for us) the aggregate RTP for a slot was proving excessively generous, there has to be compensatory algorithm in place. Secondly, how can a casino give an RTP on a brand new game which hasn't done a huge amount of spins? It would have been tested, possibly as a clone on another similar program platform already in place for another existing slot. IF a casino can give an RTP for a slot, it has to be either a theoretical one (based on the program) as you state, or a PROVEN one by using data gleaned from previous play, and ongoing therefore I think this is the 'infinite' you mention as it is open-ended. The RTP is fluid in this case. If the actual and theoretical RTP's (over many slots) are extremely close and the slots have been around for a while, it would make most people believe in the compensation theory. Simply because statistically (assuming total randomness) one or two slots would likely have paid a large series of wins and be well over and some well under. This seems seldom the case though.
Another mystery (it has happened to me several times on more than one software) is why, after playing say 30p minimum stake for hundreds of spins, getting bored and treading water that I suddenly whack the stake up to say 3.00 to rid myself of my last 18 quid, and I hit a big win. Happens often on MG. I know I'm not alone, as many winners of the big screenshots often add 'I was playing x and raised my stake to xx, when this beauty rolled in'. Not proof total of compensation, but almost seems like a form of 'forcing' which we know can be done on pub and arcade slots but shouldn't happen on totally 'random' slots. Funny old game.