AylerKupp: <Damianx> I can't tell you a win. And, as <beatgiant> and <Shams> indicated, 37...Qe6 is equal and everything else loses. Perhaps, since this was a relatively long game, it might have been the last game to finish. In that case Fischer would have been standing in front of Shifrine's board, just waiting for him to make his move. That would certainly be intimidating. So, with all variations except one losing, Shifrine might have lost his nerve. Just a guess on my part, nothing more. But I still remember Reshevsky standing in front of my board during a simultaneous exhibition and staring down while fidgeting with one of the captured pawns, clearly impatiently waiting for me to make my move. I don't remember if I was intimidated (after all, I was bigger than Reshevsky, something I couldn't say about Fischer) but I did lose that game.
But, of course, my inability to see a win doesn't mean anything so I tried several engines. Here is the eval and the top line of some of the engines I tried:
Critter 1.6a: [0.00], d=28: 37...Qe6 38.Ng8 Qxf7 39.Nxh6 Qh7 40.Ng4 Qh1+ 41.Ka2 Bd6 42.Nh2 Kb7 43.Nf1 Qh7 44.Nd2 Qg8+ 45.Kb1 Qg1+ 46.Ka2 Qg8+ and a draw by repetition.
Houdini 1.5a: [0.00], d=31: 37...Qe6 38.Qf1 Qxf7 39.Qf5 Kb7 40.Nd7 Be3+ 41.Kc2 Qxf5 42.exf5 e4 43.Nf6 c5 44.Nxe4 Kc6 45.c4 h5 46.Kd3 Bd4 47.b3 h4 48.Ke2 h3 49.Kf3 h2 50.Kg2 Be5 51.Ng5 Bf4 52.Ne4 Be5 53.Ng5 and a different draw by repetition.
Ivanhoe 946f: [0.00], d=25: 37...Qe6 38.Qf3 Qxf7 39.Qf5 Kb7 40.Nd7 Be3+ 41.Kc2 Qc4 42.Nc5+ bxc5 43.Qd7+ Kb6 44.Qd8+ Kb7 45.Qd7+ and yet another draw by repetition.
Rybka 4.1: [-0.06], d=24: 37...Qe6 38.Qf3 Qxf7 39.Qf5 Kb7 40.Nd7 Be3+ 41.Kc2 Qxf5 42.exf5 Kc7 43.Nxe5 Bf4 44.Nf3 Bd6 45.Kd3 h5 46.Ke4 Kd7 47.Ng5 Ke7 48.c4 Kf6 49.Nh7+ Kf7 50.Ng5+ Ke7 51.b3 and apparently Rybka evaluates Black as being microscopically better.
But I did see an interesting situation. Bouquet, Komodo, and Stockfish all indicated an equal or better eval for Black after 37...Qe6 38.Qf2 Qxf7. I said to myself, What!!! How can these engines miss 38...Bxf2? I checked my setup several times to make sure that I had not misplaced the pieces and I hadn't. But then I looked a little further and it seems that these 3 engines evaluated 38...Bxf2 and now either 39.f8(Q) Qxf8 40.Nd7+ and 41.Nxf8 or 39.Nd7+ Qxe7 40.Qf8+ and 41.Qxf2 as not being as good as their suggested continuation. But only a computer would consider lines like that!
For completeness, here is the eval and top line for each of these 3 engines:
Bouquet 1.6.7: [-0.09], d=21: 37...Qe6 38.Qf2 Qxf7 39.Qf5 Kb7 40.Nd7 Qxf5 41.exf5 Bd6 42.Nf6 Kc7 43.Kd2 Be7 44.Nh5 Kd7 45.Ke3 Bg5+ 46.Ke4 Bc1 47.b4 Bb2 48.bxa5 bxa5 49.Nf6+ Kd6 50.Kd3 Ba1 51.Ne4+ Ke7
Komodo 5: [-0.16], d=25: 37...Qe6 38.Qf2 Qxf7 39.Qf5 Kb7 40.Nd7 Be3+ 41.Kc2 Qxf5 42.exf5 Kc7 43.Nxe5 Bf4 44.Nf3 Kd6 45.Kd3 h5 46.Ke4 Bg3 47.Ng5 Ke7 48.Kf3 Bc7 49.Ke4 h4 50.b4 Bg3 51.bxa5 bxa5 52.c4 c5 53.Ne6 h3 54.Nxc5 h2
Stockfish 2.3.1: [0.00], d=27: 37...Qe6 38.Qf2 Qxf7 39.Qf5 Be3+ 40.Kc2 Qc4 41.Nd7+ Kb7 42.Nxe5 Qe2+ 43.Kb3