I've heard it before, that the difference between 2200 and 2500 might be just 2 or 3 key positions in a game where the 2500 can find the best move and the 2200 can't.
Looking at chess this way it's a bit discouraging that such a thin line separates players. You might have to work 5 years and/or on a few thousand more positions to squeak out the extra 2% of accuracy just to get to the next class.
Played a pretty strong player G/20 online yesterday. Saved the game and reviewed it with my computer. Now sometimes the difference between computer choice #1, #2, and #3 were significant, but I was amazed that almost all my moves were in the top 3 yet I was still ground down and lost. (I enjoyed the game, this guy was really good).
Also, in the game the ideas of the middlegame and endgame were pretty clear to me, and of course when you're following the thread so to speak it's not hard to avoid blunders.
But I'm easily below expert level... and there is really that much difference between 2000 and 2800?! It's amazing. Obviously not all my games are like this, but still improvement from 2000 and up must take incredible consistency. Top amateurs (and those below them, players like me), who are considerably worse than say, IMs, are actually a lot better than I gave them credit.