DrGridlock: Fun with Tal, and fun with computers!
It’s interesting to compare Tal’s analysis of this game (from “The Life and Games of Mikhail Tal”) with Rybka analysis of the game. Tal ‘s comments suggest that from about White’s move 13 (which he gives a “?”) onward, he is playing a superior position. To White’s move 13, Tal comments, “This queen move is clearly bad. White tries to prevent the move b5, but does so uneconomically.” To White’s move 18 g4, Tal comments, “practically forced.” To Black’s move 21 Bd4+, Tal comments, “Black spent some considerable time on this apparently natural continuation, since in the first instance the piece was sacrificed on general considerations! Black assumed that his attack, in which all his pieces are taking part, should be irresistible, and in analyzing the move 17 Bxf5, did not try to find a concrete solution.” After White’s 23 Nc3, Tal writes, “Evidently the strongest continuation, allowing White to hold the position for the moment. But now Black can quietly move his attacked pieces back, maintaining (at no cost!) all the advantages of his position.” After Black’s 25 … Nf6, Tal writes, “The excitement has died down. Material is equal, but White’s position is compromised on both flanks.” An interesting position is reached after Black’s 22 … Qg7. According to Rybka, White has three good continuations in a position that slightly favors Black. 
click for larger view1. ³ (-0.27): 23.Qa5 Rb4 24.Qxa6 Bc8 25.Qc6 Rf8 26.Ra4 Qe5 27.Rxb4 cxb4 28.Ba6 Nf6 29.Bd3 Bd7 2. ³ (-0.28): 23.Nc3 Rb4 24.Bxa6 Bd7 25.Ne2 Nf6 26.Bc4 Be5 27.Nc3 Qe7 28.Bg5 Reb8 29.Rf2 Bxc3 3. ³ (-0.28): 23.Bxa6 Qe5 24.Bc4 Nf4+ 25.Bxf4 Qxf4 26.gxf5 Be5 27.Rh1 Qg5+ 28.Kf1 Qxf5 29.Kg2 Rb4 4. ³ (-0.30): 23.Qa4 Bd7 24.Qa5 Rb4 25.Qxa6 Bc8 26.Qc6 Rf8 27.Ra4 Qe5 28.Rxb4 cxb4 29.Ba6 Nf6 5. ³ (-0.36): 23.Kh1 Bc8 24.Qa5 Rb8 25.Bxa6 Ra8 26.Qb5 Nf6 27.Bf4 Qf7 28.Bxc8 Rxa1 29.Be6 Each of these lines involves White taking Black’s a-pawn, either with White’s queen after driving off Black’s rook, or with White’s bishop, supported by the Queen. These lines contradict Tal’s comment, “maintaining (at no cost!) all the advantages of his position.” The cost to Black should have been the a-pawn, after which White’s material advantage roughly balances Black’s positional advantage. In the game, White plays his knight to c3, and instead of following Rybka’s line with 23 … Rb4, Tal plays his bishop to d7. This allows White an opportunity to completely equalize: 
click for larger view1. = (0.00): 24.Ne2 Rb4 25.Qxa6 Bxb2 26.Ra2 Bxc1 27.Rxc1 Qe5 28.gxh5 Qg5+ 2. ³ (-0.28): 24.Bd2 Qf6 25.Rae1 Rxe1 26.Bxe1 Nf4+ 27.Kh1 Qe7 28.Bg3 Qf8 29.Bxa6 Be5 30.Bc4 Rb4 It’s interesting to analyze the reason Ne2 tactically “works.” Both Black’s bishop (on f5) and knight (on h5) are attacked by White’s pawn (on g4). Tal observes that, “White cannot capture the knight – 24 g4xh5 g6xh5+; 25 Kh1 Bh3.” The discovered check from the queen down the g file, along with a mate threat after moving the bishop to h3 “saves” the bishop and knight from capture. Once the knight arrives at e2, it threatens the bishop on d4, and also makes itself available to block a check on the g-file with Ng3 – activating the threat of the pawn capture of f5 or h5. In response to those threats, Black is compelled to seek exchanges of pieces, with a resulting level position. | 
Kibitzing
(-1.64): 27.b3 Nxg4 28.Ne2 Ne3+ 29.Bxe3 Bxe3 30.Ng3 Qf4 31.Qc2 Qh6 32.Kh1 Rf8 33.Qg2 Qf4 