Ah, I was writing my reply while you posted. 14. cxd4 exd4 (threatens 15... Re8+) 15. Nc3 (Fritz says this is the only way for White to hold on) dxc3 16. Be2 cxb2 17. Rb1 Bxe2 where White's King in the open leads to perpetual.
Traxler discussion

A: 10.c3 Bg4 11.Qa4+ Nd7 12.Kxf2 Qf6+ 13.Ke1 0-0-0 14.Nf7! Qf5 15.Rf1 Nc2+ 16.Qxc2 Qxc2 17.Na3 Qe4+ 18.Kf2 This looks much better than 14.Rf1.
B: 10.c3 Bg4 11.Qa4+ Nd7 12.Nf7?? Qf6 13.Qxd7+ Kxd7 14.Nxe5+ Qxe5 15.g3 Bh3+ 16.Kxf2 Rf8+ 17.Kg1 Qe1+ 18.Bf1 Rxf1#
C: 10.Nf7 Qc5 11.d3 Bg4 12.b4 Qe7 13.Qd2 Be2+ 14.Kxf2 Ng4+ 15.Ke1 Qf6 16.Qxe2 Nxe2 17.Kxe2 Qf2+ 18.Kd1 Qxg2 19.Re1 Qf3+ 20.Kd2 Qf2+ 21.Re2 Qf4+ 1/2-1/2
I do not know the line with 10.Be2, but it is not the best move though.
9.h3 Bh4! 10.c3 Ne4 11.cxd4 exd4 12.Bb5+ Bd7 13.Bxd7+ Kxd7 14.Qg4+ Kd6 15.Qe6+ Qxe6 16.dxe6 Ng3+ 17.Kg1 Ne2+ 18.Kf1 Ng3+ with a draw
9.Be2 Bh4 10.c3 Nxe2 11.Qxe2 Bg4 12.Qb5+ Nd7 13.g3 Qf6+ 14.Kg1 Bh3 15.Qe2 0-0-0 16.d4 Rf8 17.Nbd2 e4 and black wins.
9.Be2 Bh4 10.g3 Bh3+ 11.Ke1 Ne4 12.Bb5+ Kf8 13.d3 Nxg3 14.hxg3 Bxg3+ 15.Kd2 Bf4+ 16.Kc3 Qc5+ 17.Bc4 Nb5+ 18.Kb3 Nd4+ 19.Kc3 perpetual
SirDavid, sorry about that B line, I forgot to take it out. I just found your line in A as well. I agree black seems ok there too.
Bücker has looked at C too, he examines 11.Bh4. I'll look at that shortly, I first want to further examine 14.cxd4.
SirDavid, you're right, in the 14..cxd4 line, Bxe2 before bxc1Q seems an improvement, although I can't find a clear perpetual.
In line C, I wonder what happens if white plays 12.Qd2 immediately in Conquiscador's line.

I still do not see a clear perpetual in line A, although, I do not believe white can win this position.
I just realized what I meant to bring up in line B: Nf7 on the 11th move, rather than the 12th:
10.c3 Bg4 11.Nf7
After 11...Qb6 12.Qa4+ Nd7 is no longer possible.
I'm losing track of everything here. Anyways, mathijs, that's the problem I'm having too- 12. Qd2. Rybka is better at these things- can anyone run the position after 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 Bc5 5.Nxf7 Bxf2+ 6.Kf1 Qe7 7.Nxh8 d5 8.exd5 Nd4 9.c3 Bg4 10.Nf7 Qc5 11.d3 Bg4 12.Qd2 through Rybka?

14... Nc5 15. cxd4 Nxa4 16. Nxd8 Qh4+ 17. g3 Qxd8 18. d5 b5 19.h3 bxc4 20.hxg3 Qxd5 21.Rf1 Qg2 22.Rf8+ Kb7 23.Nc3 Nc5!! and black is winning positionally.
Mathijs, 11.Nf7 Qd7 12.Nxe5 Qf5 13. Bd3 Qxe5 14. cxd4 Qxd4 15. Bb5+ Kf8 is evidently equal. I'm turning my brain off now because it's tired, and pretty much everything is Fritz. EDIT: After thinking more, it says Black is ahead by over a pawn.
I just found out mr.Jerz, who was referenced here earlier, already looked at both our problem lines: here
Variations B and C are exactly the lines we were having trouble with, i.e. 10.c3 Bg4 11.Nf7 and 10.Nf7 respectively. So, it loooks like black is ok in the Nxf7 lines.
It just goes to show how insane this variation is that it took us 8 pages to figure out the FIRST HALF of the variation. Bxf7+ I believe may be OK for Black because he only needs compensation for a pawn, not a Rook. We may as well try to figure out everything anyways, though. I don't think I'll be here for the whole discussion because of my vacation.
OK...
mathijs, A: 14... Nc5 15. cxd4 Nxa4 16. Nxd8 Qh4+ 17. g3 Qxd8 18. d5 b5 and Fritz still insists there's perpetual.
B: 12. Nf7 Qf6 with nothing useful against Bh4+
C: 10. Nf7 is really hard to meet... The best I can do so far is 10... Qc5 11. d3 Bg4 12. Qd2
Conquiscador, I guess we're not ready yet. I'm stuck on this line. In the Bxf7+ line you gave with 9. Nf3, I think you should consider 8...Bg4 9. Qe1 bxc6 as well. After a while, everything seems to dissapate into a slight White advantage.