Why don't people do this more often?

It's just not usual to have the chance. 5. f3 is a strange move.

White gets the bishop pair when you do this so there has to be some kind of compensation. I think the bishop was more dangerous sitting on c5 because it's hard for White to do anything about it. 

After 6...Bxg1 7.Rxg1 you probably need to get at White's king before he can develop and get coordinated. Maybe you can do this with 8...h6. After 9.Bxf6 Qxf6 you might be able to play ...d5 next turn sacking a pawn and opening lines. If 9.Bh4 then 9...g5 followed by 10...d5 could be similar but you might sac your c7 pawn instead. White has to be careful with ...Qh4+ and ...Re8 threats in the position. 6...d6 probably lets White get organised.

In the game White gave the bishop pair back and was left with a king in the center.

Why is f3 so bad?   It protects the e4 pawn, and it prevents ..Kg4

Because it opens two dangerous diagonals near the uncastled king and weakens the protection of the g3 square, which is an important square near the king. 

It weakens the king's position and weakens the dark squares. After you played ...Bc5 in the game White couldn't put anything on that entire g1-a7 diagonal because your bishop practically owned it. Castling would be difficult to pull off and White would also have to be careful that ...Qh4+ wasn't a threat. 

If White tried something like Qe2 and Be3 to get rid of the bishop this would cost White two turns giving you time to develop and prepare some kind of attack with something like castling, playing ...d5 and getting your rook to e8. You only want to play a move like f3 when you need to and you're sure there aren't any tactical problems. 

Here are some games with tactics against f2 and f7:

http://www.chessgames.com/perl/chesscollection?cid=1006294

You don't have to watch all of them because there are probably 70 games. But the first games can be useful to watch. They start out with very simple attacks and become more complicated.

I would say that taking the knight wasn't that good because the bishop is delevoped while the knight wasn't and he can still castle queenside.

One more reason not to exchange B for N is the MINOR EXCHANGE.

        To put it mildly.

pawnpatrol, i think that is a good simple answer, easy to understand, not some long article i'd be bored to even look at lelz.

More importantly, it loses a tempo to develop (Bd3/Nc3 instead) and takes the f3 square from the white knight.

Notice having your bishop on c5 prevents castling whether you trade it for the knight or not.

However, not making the trade has the additional benefits of pressuring white's weakened dark squares on the king side as well as presenting white the problem of where / how to develop his king's knight.

Not to mention by trading a developed piece for an undeveloped piece you're losing a tempo in the opening.  Also someone mentioned losing the bishop pair (although that won't be relevant to your results for well over 1000 rating points).

Bishop = 3.1; knight = 2.9

That is quite an exaggeration.

If I were to believe my experience, bishop vs. knight would be about equal in endgame, bishop>knight in middlegame by about half a pawn, and bishop pair varies from zero to more than a pawn (all that depending on the position of course).

If I were to believe Kaufmann's statistical evaluation (here), the bishop is essentially equal to the knight but the pair of bishop is worth about a half-pawn.

But there is no way than, say, 2B vs. R+N is an equal fight.