Since I don't have oceans of time, I haven't had it look at positions longer than say 10 or 20 minutes for the more complex ones.
What is interesting is that this seems to be a game and especially from move 16 onwards that demonstrates the horizon effect of the engine, which is by the way Rybka 3, only I run it with the Fritz program, I think (never know quite how that works).
Of course 16. Bxd6 is a stunning move. The moves that follow this move are pretty much in the top choices by the box, although it evaluates the position still as equal.
In the following position
attention was given on chessbase to the move Bxc4, instead of Kg7 which was played ingame, pointing out that it is no good either.
The engine initially disagrees, of course, but after playing out some of the variations it prefers, it changes its opinion drastically regarding the evaluation.
Two interesting lines:
Beautiful game. Worth analyzing.
Edits: pgn things...