I would never do it up by 3.
The only time I think it's the higher percentage play is when you're tied or losing by less than 3 and your opponent's timeout situation is such that you can basically wind the clock down until the FG try.
Look at it from the offense's position. Imagine the Pats are in a tie game with 2 minutes left. First down at the 2. Opponent has no time outs. Would you rather: (1) kneel on the ball 3 times and kick a field goal with time expiring, or (2) go for the TD.
I pick number 1. You win the game over 90 percent of the time in that situation. I don't think the same is true of scoring the TD with 2:00, although it also depends on the offense of the team you're playing.
If you change the scenario just slightly -- same exactly situation except you're down by 1 it's a much harder call. When you're losing, the tendency is to want to take as many shots at getting points as you can. But actually, burning the clock to the last seconds and then trying a winning field goal is actually SMARTER when you're down 1 (or even moreso 2). The reason is that your odds of kicking a chip shot FG are much higher than your odds of holding your opponent from scoring a winning TD in two minutes. It's pretty hard to be that gutsy, but if we're talking percentages, it's the higher percentage play, and a great coach is the coach who picks the option that wins more than not.
Once you start moving the field goal out a few yards, put weather into the equation, or give your opponent a hypothetical time out or something, everything changes.
(One other way to play it is burn up clock with two plays -- about 1:25 -- then try the TD on third down. If you make it, you're up by more than an FG and your opponent has only about 20 to 25 seconds and no time outs. If you don't make it, you can kick -- just gives you two chances to score.)