Note that the GE terrain is just a more or less accurate approximation of the real terrain (in fact, any digital data about terrain is just an approximation since the accuracy depends on the denseness of the data collected).
Now just like with the variation of imagery quality al around the world, the terrain quality is not the same everywhere either. Smaller or sometimes even bigger terrain differences (especially when these are only at a very local level) often "averaged" and you can only tell that the overall accuracy is more or less okay throughout a couple of miles (and not the details within this distance).
Certainly falling into an "averaged" area where the slope goes exactly the other direction, is a pretty much unfortunate thing. Is the difference very big? Would it prevent you from modelling something completely?