Using 'tactics trainer', you win a few moves after your clever sacrifice or after capturing the opposing queen etc. In a lot of cases, after the moment of brilliance, the game is completely over. However, you won't get maximum points for, let's say a checkmate in 5 if there was a checkmate in 4 available. Of course quickest is best but we are supposed to be training our tactics; what training is going on (after the brilliant sacrifice) when you've got two queens and they've got a pawn?

Quickest is usually best. You could be low on time, or close to the guillotine limit e.t.c; and it's always best to practice the quickest way to win.

GMs don't always pick the easiest win in their games, but they're good enough to be able to do that thing whenever they want.

The only training benefit would be again to hone your vision. In a real game however, the quickest checkmate is often not played intentionally because it might involve unnecessary risk. For example, you might be in a rook and pawns endgame where you can technically win in 13 moves with a complicated combination involving risky checks for both sides, or you simply force a trade off of rooks and win in 20 moves in a blatantly easy endgame. Even GMs will take the easier and safer route regardless of it being an extra couple of moves before winning.