It is mentioned in one of the papers (or possibly the doctoral thesis, which is nice as it gives an overview of the whole chain of improvements) that it can still happen but that it is much better compared to what they benchmarked it against (boehm, iirc). Hard to tell if it delivers what it promises though..