I can believe that Rybka could Not find a defense for White. But White did, in fact, have a possible defense! ********* It would have required White to anticipate the key elements of the follow up attack by Black. So White needed to react after Black had been committed to the attack (through Qh4, Bxh3, and Nh4) and before Rg8 had been played. This is not likely unless he already knew beforehand the result of Black's attack, because of the seeming material advantage that White had at the time. White would have to know that the edge was in favor of Black before he would have or could have possibly looked for a drawing strategy. ******** Yes!!!! White needed to play for the draw early enough in the attack and he could have stopped Black from winning outright. ******** Computers generally look for wins; they are not good at finding forced or likely drawing lines. Generally speaking they only fall into those lines by accident and circumstance, not by design. 30. Kf2 as you mentioned was a better choice for White. An example variation starting with that move would be 30. Kf2 Rg8 31. Ke1!!! ...and now gxf1Q or B or R it does not matter...32. Bxf1 Nxf3+ 33. Ke2 Qh1 and the attack is basically stopped in it's tracks. The position would be basically dead even at that point.

This position looks lost. The N on a6, B on b2 are spectators. Well played black, however I feel White was the author of his own downfall.

Possibly 32...Qg3ch is also playable, with much the same idea as in the game. A sample line: 33.Ke2 Nxf3 34.Nbxd6 (say, as in the game) ...Bh4 (threatening ...Qf2#) 35.Bxe5ch Nxe5 36.Kd2 (running for it, but what else is there?) 36...Qxd3ch 37.Kc1 Qxd1ch 38.Kxd1 Nxc4 39.Nxc4 f3 and those advanced pawns ought to be worth more than the exchange. Instead of 34...Bh4, better might simply be to recapture on d6 (24...Nxd6 25.Nxd6 Bh4 26.Bxe5ch? Nxe5 etc, say). ...Rg3 seems to add weight to the attack, but ...Qg3 looks quicker.