Counterintuitive

The first diagram shows another interesting moment from the game Jakovenko-Illescas, Pamplona 2006. Black, who is under some pressure, has just played 18...Rd8. The idea seems clear enough: he wants to play ...b7-b6 and develop his queenside without allowing the reply Be4 followed by Nc6, so he pins the bishop against White's queen.

White had previously sacrificed a pawn to reach this position and must now continue to work with threats:
19.Qd2 b6!?
At first sight a surprising move. Instead of protecting his h-pawn, Black carries on with the plan of developing his queenside. But it turns out he didn't have a lot of choice, for example: 19...Kg7 (worse is 19...Kh7 20.Rad1! intending 21.Bxg6+, when Black has nothing better than 20...Kg7) 20.Qa5!? Qe7 21.Qc3 (threatening 22.Re3 followed by 23.Rf3) 21...Rd4 and now 22.b4! is very good for White.
20.Qxh6?!
Why am I awarding the dubious mark to this natural move? Well, it's not because it isn't a good move – it's because there seems to be a better one available. As Max Notkin reported in Chess Today, the leading computer programs have a strong preference for 20.b4! here. Without contacting Dmitry Jakovenko we'll never know if he considered 20.b4 and rejected it or failed to consider it altogether. I tend to believe the latter because even a cursory analysis reveals a dramatic increase in Black's difficulties. After 20...Bb7 21.bxc5 bxc5 22.Rab1 Rb8 White shifts the action back to the kingside with 23.Qxh6! and if 23...Ne8 then 24.Rb6! sets up very powerful threats. (See the second diagram. As Bent Larsen once expressed it, all pieces are attacking!) Going back, if Black captures the upstart with 20...cxb4 (instead of playing 20...Bb7) then 21.Qxb4 breaks the pin on the bishop, attacks the pawn on b6, and also threatens to snag the exchange with 22.Be4. This too is very strong for White. I'm speculating, but it seems to me there is a psychological reason why even a strong player could miss the move 20.b4. With this pawn configuration, setting up a lever on the c-pawn is not a common strategic device because of the reply ...cxb4, when if anyone's pawn position has been weakened, it is White's. So in the search for candidate moves 20.b4 would tend to be dismissed almost subconsciously. It's an interesting theory – if Jakovenko annotates the game somewhere we'll probably find out!

After 20.Qxh6 play continued
20...Bb7 21.Re3 Qg7 22.Qh4 Rd4 23.f4
and now Max Notkin suggests the following line for Black:
23...g5!? 24.Qxg5 f6 25.Qxg7+ Kxg7 26.Rg3+ Kf8 27.Ng6+ Kf7
with very good compensation for the pawn.