There are a number of moves that can be played after 1.e4 c5 2.Nf3; the three commonest are 2...d6, 2...e6 and 2...Nc6, but there are also a number of rarer moves that occasionally crop up. One of these is 2...a6, the O'Kelly variation.
This move is not as popular as the other three moves because black is showing his hand first; he's committing himself to a move which may not be useful against the set-ups adopted by the white player. It's no co-incidence that one of white's commonest responses here is 3.c3, attempting to show that the two interposed moves are in white's favour. Other moves along the same line of thought are the King's Indian Attack, 3.d3, and the gambit 3.b4!? with which I once beat FM Michael Franklin. Another popular option for white is to play 3.c4, aiming for an Open Sicilian with a Maroczy Bind, and this is likely to transpose into some line of the Kan.
All of these, except possibly my gambit line, are good ideas, and have their points in favour. What is not a good idea is to carry on on Open Sicilian auto-pilot and play 3.d4?!, as many players do when faced with this variation for the first time. The reason soon becomes clear: after 3...cxd4 4.Nxd4 Nf6 5.Nc3 e5 (diagram), black has got his ideal Najdorf set-up without the concessions made by having to play ...d6: his bishop can freely roam to c5 or b4, and his pawn can come straight to d5 in one go (as it would, for example, if white were to play 6.Nf5 here). (Note the contrast with the Lowenthal and Sveshnikov variations: in both of those, an early ...e5 can be met by Nb5, and white's attack on the d6 square makes it that much harder for black to get a good position out of the opening.)
Both these themes came into effect quickly in the game Hale-Owens from this season's 4NCL; from the diagram position, the game continued 6.Nb3 Bb4 7.Bd2 O-O 8.Bd3 Re8 9.O-O d5, reaching our third diagram. White could probably have equalized by swapping off a pair of pieces here with 10.Nxd5 Nxd5 11.exd5 Bxd2 12.Nxd2 (not 12.Qxd2?? e4) Qxd5; the interesting question here is whether 13.Qh5 can cause black enough hassle to make up for the slight positional advantage the pawn structure gives her.
As it happens, white played 10.Re1, to which black replied 10...d4, and from then on, it was clear that the advantage was with black. It's said of the Sicilian that black is usually doing well if ...d5 can be got in with no ill-consequences, and this certainly proved to be the case here.