I think it's the way in which they identify likely voters (which is propreitary)....
For instance, Gallup had President Bush's numbers way higher than all other pollsters back in late July and August because supposedly they way they identify likely voters is favorable to Republicans. ARG is just the opposite.
Kind of the same thing how some pollsters push leaners differently, and how that really affects their numbers (i.e. Rasmussen and Zogby getting different results)
I suspect you are right. It's either their LV model or possibly how they "randomly" select people. It's possible they oversample the inner city telephone exchanges or something where Republicans are not as solid.