The point I was making is that whether or not something, especially a more complex algo like MTP, is memory-hard in practice can obviously depend on all sorts of details of the algorithm. And once such details have been spotted and fixed, I can't see how it would not be 'fundamentally' memory-hard.
Pretty vague.
Well I'm not a Khovratovich or Dinur or Nadler, but if Khovratovich has been in contact with the two and he says the current fix removes the attack vector, then it's reasonable to believe that.