The point is that 39 ♕x♗ does not work due to 39 ... ♕g6+! 40 ♕x♕ d1=♕+!, this zwischenschach (<in-between check>) winning easily after 41 ♔b2 px♕. The zwischenschach 40 ... d1=♕+! is what Sadvakasov missed when he played 38 ♕g3?.

(24...Bxf6 25.Bf4 Be5 26.Bxe5 is not better for Black) ---------------- - - - - - - - - - -
31.Nxc8+ Rxc8 32.Rxc8 Qxc8 since d4 is protected by N/f5, which in turn is guarded by Q/c8.

********

click for larger view

35...Ka7-a8! was necessary to avoid 36.Nc6+ <double check!> Ka8 37.Qa7#

********
After 36.Qg3 Qg6 is best (with 37...d2+ threatened and White cannot reply 38.Qxg6 since ...d1=Q#). So 37.Qxg6 hxg6 gives Black the endgame advantage. 38.Nxe6 fxe6 39.Kc1 e5 40.Kd2 e4 wins .

The following variations show how easy Black can err.

36...Qh6 loses d3-pawn

36...Qd2 37.Qg8+ Ka7 38.Qg1! sets up mate threat as before.

36...Qxg3 37.hxg3 Ka7 <37...Bg4 38.Kc1 and Black could lose the d-pawn)> 38.b3 Kb6 39.Kc1 again.

Morozevich vs M Vachier-Lagrave, 2009

where MVL wins by "not cracking" under pressure.

I suppose the point is that 22. Nc5 puts three white pieces en prise - the knight and both bishops. So it doesn't matter whether black takes one of the pieces now (with 22...Qxc5) or ducks his king into safety and looks to take one of the offered white pieces later. White cannot extricate all three of his en prise pieces in hurry.

On balance, 22...Kc7 makes pragmatic sense. Get the king into safety first and grab material second.

Incidentally, after 23. f6, Fritzie prefers 23...Qxf6 when white still has three pieces dangling. That evals at -3.6, where exd3 is just -1.67.

Robbolito gives, at depth 18, ...Qxc5 as the best option:

22...Qxc5 23.Bxe4 Kc7 24.Qf3 Qa5 25.Bf4+ Kb6 26.Qb3+ Ka7 27.Kd1 Bh4 28.c3 Bd7 29.Qxf7 Rxg8 30.Qxg8 Qb5 31.Qg4 Bf6 32.Qe2 Qxe2+ 33.Kxe2 dxc3 34.bxc3 Bxc3

(-2.38) Depth: 18/45 00:00:32 18984kN

But after it changes to ...Kc7

22...Kc7 23.Qg3 Qxg3 24.hxg3 exd3 25.Nxd3 Kd6 26.g4 b5 27.Rg7 f6 28.Rxh7 Bd7 29.Bf4+ Kd5 30.Bc7 Re8 31.a3 Ke4 32.Kd2 Kf3 33.Rg7 Bc8 34.Rg6 Ke4 35.Bg3 Rh8 36.Ke2

(-2.32) Depth: 22/52 00:08:15 298mN

And this is both lines with Rybka 4 at depth 18:

1: Darmen Sadvakasov - Maxime Vachier-Lagrave, Aeroflot Open 2006

click for larger view

Analysis by Deep Rybka 4 w32:

1. (-2.26): 22...Kc7 23.Qg3 Qxg3 24.hxg3 exd3 25.Nxd3 Kd6 26.g4 b5 27.Rh8 Bb7 28.Rxh7 Rg8 29.Rxf7 Rxg4 30.Bf4+ Kd5 31.b4 Bc8 32.a3 Ke4 33.f6 Bd8 34.Bc7 Be6 35.Nc5+ Kf5

2. (-2.07): 22...Qxc5 23.Bxe4 Kc7 24.Kb1 Kb6 25.Qf3 Ne5 26.Qb3+ Qb5

And the human (mine) explanation:

Is a matter of initiative, which is a very important thing to have in those positions: black already has a handy piece capture (exBd3) which can also contribute in his attack, while after Qxc5 the Bxe4 capture leaves Qc5 misplaced for the attack (as c2 is already covered by B and Q) and, furthermore, he has to move the K anyway (to cover it from checks from, esentially, every white piece), so you end up losing tempi with the Qxc5 capture.

For 22.xx, Bxe4 was also possible, but maybe Sadvakasov (being already lost) wanted to tangle things a bit to see if Maxime got it wrong under pressure/complexity.

22... Qxc5 is perfectly playable and winning according to the silicon but the game's 22...Kc7 is good too. But either my silicon is messed up or MVL missed a few tactics. He went from clearly better (+2-2.5) to about equal after (24. Ne4, Qe5, 25 Qg3+!)

In response to 23.f6, Black could've played the very non-human Qxf6! Three of white's pieces are hanging, but he can save all of them. 24. Bg5, Qe5 25. Nxe4 seemly holds the white camp intact, but Black has a final sting 25... Bh3! winning the exchange and game.