In this position black just played 19...Qd1-b6 and offered a draw. While looking over the game with Stockfish 7 (64 POPCNT) it was showing a winning advantage for white of about 2.5 Pawns while Komodo 8 was evaluating the position at 0.00.
I decided to run some Shootouts and with Stockfish white scored 5 wins out of 5 while Komodo scored +1 -0 =4 for white.
The curious thing was that playing white with Stockfish and black with Komodo 8, Komodo shows things equal until you actually start making moves. Some further experimentation resulted in almost every case that after a few moves, Komodo's evaluation started making a sudden jump to a huge advantage for white. Also, the two engines disagreed on almost every move!
I also checked the position with some other engines:
Houdini 1.5 - evaluation 0.00
Fritz 12 – evaluation 0.00
Gull 3 - evaluation 1.57
Stockfish 7 64 – evaluation 1.85
There are some basic differences in the engines: Stockfish is a rather deep searcher, but it does it through aggressive pruning of the tree of analysis. The advantage to this is quick analytical sight and tactical ingenuity, but there are also some disadvantages. Stockfish can miss some resources hidden deep in the position. Komodo 8 is slightly faster in its evaluation has been fine-tuned by GM Larry Kaufman and it seems to be slightly more attuned to positional nuances especially when it comes to evaluating positions that have a material imbalance.
What's the correct evaluation? Honestly, I do not know, but would be interested in seeing if anybody cares to test the position. In any case, in an engine assisted correspondence game, I'd play Stockfish's 20.Nd3 just to see what happens because it seems white has almost no chances of losing.