|zanzibar: <zydeco> that's correct, but theory is a very demanding taskmaster.|
<MillBase> has 119 games with 10...dxe5 and White scoring 55.4%,
whereas it has 67 games with 10...h6 scoring 49.2%.
That suggests 10...h6 is quite playable.
And if one looks at SF8's eval of 10...Nfd7 (<MB>'s 3rd candidate move), it's a perfect draw at 0.00.
* * * * *
Playing over this game, it looks like 13...Nbc6 was not precise, with 13...d5 giving the better eval. Allowing the engine to dig deeper it looks like it does indeed hold.
(If Black played 13...d5)
click for larger view
rnb1kb1r /1p4pp/p3p3/3pn1B1/3N4/q1N5/P1PQB1PP/1R2K2R w - - 0 14
One nice thing about 13...d5 is that it prevents Ne4, delaying the entry of White's QN from activating. But playing it over, it's clear that White gets a ferocious attack, which pushes Black to the wall.
Aside - how does White get the QN into the game after 13...d5. By sacrificing it on d5! (see 16.Nxd5 below)
[Stockfish 8 64] 28:+0.67 13...d5 14.O-O Bc5 15.Kh1 Bxd4 16.Nxd5 exd5 17.Qxd4 Qd6 18.Bh5+ g6 19.Rf6 Nbc6 20.Rxd6 Nxd4 21.Rd8+ Kf7 22.Rxh8 gxh5 23.Rf1+ Nf5 24.Rxf5+ Bxf5 25.Rxa8 Bxc2 26.Ra7 Kg6 27.Bf4 Kf5 28.Bc1 Nd3 29.Bd2 d4 30.Rxb7 Nf2+ 31.Kg1 Ne4 32.Ba5 d3 33.Rf7+ Kg5 34.Rxh7