Most opening manuals give 13.Nh5 as an improvement for White.
The manuals continue with 13...Nxh5 14.Qxh5 O-O, and now David Levy gives 15.Bxe7 Qxe7 16.Qe2 as roughly equal.
The exchange sacrifice 13...Rxc3 is mentioned nowhere except in Michael Stean's Sicilian Najdorf, and there it is given ?! and called "dubious" because of 14.Nxg7+ Kf8 15.Bh6! Kg8 16.bxc3 Bf8 17.Ne6!.
Nevertheless, in my game against Tim Hall (USCF 2100) at the Somerset NJ Quads in September 1992, I played 13...Rxc3! anyway.
Over the board, I found a simple refutation to Steans's 14.Nxg7+ ?! Kf8 15.Bh6 with 15...Re3+!.
Now 16.Bxe3 Kxg7 is clearly in Black's favor. And both 16.Kf2 and 16.Kd2 allow 16...Nxe4+ when 17.Kxe3?? Qb6+ forces White to give up his queen or be mated.
Also, after 16...Nxe4+, other king moves (such as 17.Kf1 or 17.Kc1,depending on White's 16th move) allow 17...Qb6 which removes the black queen from danger by discovered checks while simultaneously protecting the rook on e3. The same holds true after the immediate 16.Kf1 Qb6.
Consequently, Hall recaptured with 14.bxc3.
After some inexact play by both players, Black won in 40 moves.
{This article originally appeared in the November-December 1992 issue of Atlantic Chess News}