Obviously if you get checkmate from it it's fine, but at what point does it become unacceptable?
If you can't calculate a clear win after sacrificing your queen, do not do it.
There are (basically) two ways to justify a sacrifice: checkmate, or win the material back.
When you sac a queen you are probably not winning the material back unless it's part of a little combination, like setting up a knight fork (although it is possible, especially if you have a far-advanced pawn), so generally you won't do it unless you can mate your opponent as a result.
(I assume that you are talking about "real" queen sacrifices and not something like trading your queen for two rooks.)
It's the queen, the best piece on the board. Sacrificing it is almost always bad!
I think most people never play a correct queen sacrifice in their lives, apart from those cases where it checkmates immediately or the threats are so large that it wins the material back immediately.
There are cases where a pure positional queen sacrifice seems to work, for instance there are some lines in the King's Indian where Black sacrifices the queen for two pieces and two pawns, which is not usually considered equivalent to a queen. This is known as Bronstein's queen sacrifice line:
[FEN ""] 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.f3 O-O 6.Be3 e5 7.d5 Nh5 8.Qd2 Qh4+ 9.g3 Nxg3 10.Qf2 Nxf1 11.Qxh4 Nxe3
Black will also win the pawn on c4. I don't know what current theory thinks of this line, but I don't think it's refuted.
But, again, there's a reason this line is famous -- positional queen sacrifices almost never work.