rwbean: It's so much fun going over old games with modern hardware and software! I vividly remember reading about this game in the local Sunday Mail newspaper in Brisbane. It was a "How to beat a GM" article. Also I had read about this opening trap in Barden's "Play Better Chess". Now with Rybka 2.1o and a dual P4-3.2GHz with 2GB RAM, I can "see" an awful lot more. I will still look back on this comment in five years' time and laugh though!
There are five reasonable moves on move 5: Nxc3, Bb4, Qh4+, Nc5, and Ng5. Bb4 is good enough to draw and Qh4+ is supposed to be inferior. But even with Qh4+ Black retains the option of 6... Nxc3 so I don't agree.
On move 6, Black can still play 6... Nxc3 7 bxc3 Qa4 =. 6...Nxg3 may be slightly inferior.
Rybka thinks 8... Nxh1 is slightly inferior: its suggested continuation is 8... Bg4 9 Bg2 Nxh1 10 Nxc7+ Kd7 11 Nxa8 Nc6 12 Be3 Be7 13 Qe2 Rxa8 14 o-o-o Ke8 15 d4 (evaluation -0.15). I think the point is that if White tries 12 d4 in this variation, Black really can play 12... Bxf3 in reply, because there is no winning Bg5+ for White.
In the game 12... Bxf3? is a big mistake. Rybka suggests 12... Kb8 13 Be3 Kb8 (evaluation -0.27). After that Black is in a very difficult position - White's game plays itself and it's no big surprise Black made a final blunder with 17... Kb8? instead of 17... h6. But the game was lost by then anyway.