I can only remember two cases where I had to turn on the full-tail-call option to the Kawa compiler, and even then performance wasn't a problem. The SISC Scheme interpreter does reasonably well too, and it has full tail call optimization all the time.