In this game between Guseinov (2622) vs. Bauer (2634):
Crete 2007 · French, Winawer, Alekhine (Maroczy) gambit (C15) · ½-½
[FEN ""]
1. e4 e6
2. d4 d5
3. Nc3 Bb4
4. Nge2 Nc6
5. a3 Ba5
6. b4 Bb6
7. Na4 dxe4
After dxe4, Guseinov played 8. Bb2.
Why did he not take the black dark square bishop on the b6 square with his knight on a4?
There is no hurry. After 8. Bb2 the bishop on b6 is not going anywhere. NxB continues to be available to white until black does something about it like a6. That means that delaying NxB gains a tempo if black has to make a less useful move to try and "save" the bishop like a6. Since recapturing with the a pawn probably gives black a better game it is worth waiting to see if he will throw away this option by playing a6.
Delaying the capture also retains the possibility of playing Nc5 which may be stronger in some circumstances.
Finally an immediate NxB will be met by axN opening the a file for black and allowing the a8 rook to pressure the now weak, backward a3 pawn.
So, in short, delaying the capture retains flexibility. That way white has more and possibly better options. This "maintaining the tension" is one of the things which separates weak and strong players. Weak players are in too much of a hurry to release the tension in a position. Strong players maintain the tension and with it give their opponents more opportunities to go wrong.
One way of thinking in these situations in the opening is the following: while white gains the bishop pair after taking on b6, it also opens up the a-file for black's rook. Moreover (as Brian Towers pointed out in his answer), depending on the situation, white may actually want to be able to place a knight on c5 in some lines down the road. So white doesn't know whether he wants to take the b6 bishop yet.
Meanwhile, white knows that he will have to develop the c1 bishop sooner or later, and b2 is not a bad square at all for that bishop, where it has great potential for activity later in the game, giving white ideas of making a d5 push.
Therefore it may be a good idea for white to make the obviously good move (which he is very likely to make at some point in the game anyway), and only afterwards consider the move that he's not completely sure about yet. In this particular example it becomes even more rational to do it this way, since the option of playin Nxb6 doesn't go away.
The bishop's more out of play on b6 than White's knight is on a4. Also, White could play c4 (threatening to trap the bishop with c5), after which Black may be forced to give the bishop breathing room with ...a6. Then White can play Nxb6 since ...cxb6 will be forced, which is less ideal for Black than recapturing with ...axb6.
Actually, the rest of the game did not go well for White. It seems to me that Na4 was a bit optimistic. White was hoping to play a gambit but does not really have enough initiative to justify playing on a pawn down, so if he does give up a pawn he must try to regain it. Putting the N on a4 gives the pawn but makes it harder to regain. Black then grabs the pawn and White then neither regains it nor develops any initiative. As various people have said, Bb2 does have its merits, but perhaps only as the better prospect in a position already a bit inferior.
After 8. Nxb6 axb6, white is more or less forced to play 9. Bb2 anyway. With the bishop gone, d4 is no longer hanging, but after other moves, 9. ... Nxb4 could be a threat due to the pin of the a-pawn (10. axb4 Rxa1). (Of course 9. c3 would also prevent Nxb4, but that is clearly inferior) After 9. Bb2, Black will make his 9th move, which could be 9. ... Nf6, 9. ... Nce7, 9. ... Nge7 etc. (though the last one looks quite bad now, because the knight on c6 has nowhere to go).
If white plays 8. Bb2 first, black of course cannot reply 8. ... axb6. He can also not prevent 9. Nxb6, so he will make another move. 8. ... Nge7 and 8. ... Nf6 look reasonable, possibly also 8. Nce7 (although d5 might be an idea for white - however as you can see these are the same moves black would have to consider on his 9th move if white took right away). Anyway white can now see how black continues, and then decide whether or not to take the bishop on the next move. In the game, 8. ... Nf6 happened, after which white decided not to take the bishop. This is reasonable, because taking the bishop right away is better after Nge7. He didn't even take after 9. c4 a6, although 10. Nxb6 axb6, as pointed out by Brian, is probably an option. After 10. g3 however he got a very good game due to the useless black bishops, but on his 13th move, white blundered with a strange rook move (13. Ra2??) that doesn't seem to make any sense (13. Nac3 followed by Nxe4 would have been the correct idea, or 13. c5 possibly followed by 14. Nac3), and was worse, but later managed to escape into a drawn ending (it is good to know that rook+rook pawn + bishop pawn (on the same wing) vs. rook is usually a draw!)
