From what I understand of it, one weakness is that I think it uses the FTR as one parameter in the decision to raise the target. Basically, in my understanding, it won't raise the target if doing so would breach FTR - and that should operate as a safety valve to prevent a long line that's capable of, say, a ssteady 2mbps at a 6dB marging being reduced tenfold to 0.2mbps with a 15dB margin (assuming 600kbps for each 3dB).
However, FTR isn't ever re-calculated after the initial training period, despite the fact that the initial FTR becomes innapropriate after improvements, such as filterred sockets, better routers, etc. I do strongly suspect that's part of my own specific problem, since my typical speeds have improved at least three-fold since my iniital training, thanks mainly to chopping off the bell wire and getting a better router.
As always, I have to stress that's only my understanding, based on what I've dug up form my own research. I could be wrong.