A trap in the King's Indian?

The best-scoring line against the King's Indian Defense is the flexible 5.h3! 0-0 6.Bg5! In Mega Database 2013, White scores a gaudy 64.6% in 2938 games! That is better than he does against garbage like the Englund Gambit, where he only scores 57.7% in 1280 games in the main line (1.d4 e5? 2.dxe5 Nc6 3.Nf3 Qe7) - although White does get 66.8% with 4.Qd5!, almost forcing Black to make his pawn sac permanent with 4...f6, when White's best-scoring line is 5.exf6 Nxf6 6.Qb3 d5 7.Bg5! (82.7%, but with a tiny sample size - only 27 games). But I digress.

The 5.h3 0-0 6.Bg5 line also sets a little trap: Black's most natural and thematic move, 6...e5, is here a blunder, dropping a pawn to 7.dxe5 dxe5 8.Qxd8 Rxd8 9.Nd5! GM Yury Shulman in his lecture at the North Shore Chess Center last year discussed a 1996 game in which he fell into this trap. Though disgusted with himself, he played on as though nothing had happened. He went on to roll his 2460-rated opponent like a joint, as NM Marvin Dandridge would say:

Where did White go wrong? According to Houdini 3, the natural 11.Nxc7 was already a small inaccuracy, and 12.Nd5 a more serious one. After 12...h6! White had to surrender his dark-squared bishop for Black's knight. This is a big positional concession, as Black's unopposed dark-squared bishop will be a holy terror once it sets up camp on c5. Black already had sufficient compensation for his accidentally sacrificed pawn, and White's game went downhill from there. (Note that all numerical assessments in this and the following game are by Houdini 3.)

How should White play? As I say, Houdini says that 11.Nxc7 is already a little inaccurate. But 11.0-0-0, White's most successful move (3-0 in Mega 2013), is actually a blunder. As I indicate in my notes to the above game, Houdini then analyzes 10...Rf8 11. Nxc7 Nxe4! as leading to equality. White's best move is instead 11.Rd1!, threatening to win a piece with 12.Nxf6+. Note that unlike the similar 11.0-0-0 it does not leave f2 vulnerable to a knight fork. After 11.Rd1!, Black must lose a tempo with 11...Rf8, and now 12.Nxc7 Rb8 13.f3! leaves White with an advantage that Houdini assesses as +1.32.

This line is seen in the following game. Incidentally, the note to Black's 9th move is very interesting. Houdini considers 9...Nbd7, dropping the c-pawn and leading to a position where White has a +1.32 advantage, inferior to surrendering the exchange with either 9...Nxd5 10.Bxd8 Nf6 (+0.94) or 9...Rxd5 10.cxd5 Ne4 (+1.09).