Here's another example each of a rook promotion and a bishop promotion from actual play. Neither specific promotion was quite absolutely necessary (you'll see what I mean), but neither one was at all a superfluous underpromotion either.
In the game Sajtar-Benko (Budapest 1954), the following position was reached after
[FEN "8/5KP1/7k/8/6P1/8/8/8 w - - 0 1"]
1.g8=R! (1.g8=Q??=) (1.g8=B?? Kg5=) (1.g8=N+ Kg5 2.Nf6 $18) (1.g5+ Kxg5 2.g8=Q+ $18) Kh7 2.Rg6 Kh8 3.Rh6#
Here promotion to a queen is stalemate, while promotion to a bishop allows Black to capture the remaining pawn with a draw by insufficient material. So the only promotions that win here are
86.g8=N+ (which is a long, drawn out win after
86...Kg5 87.Nf6) and
86.g8=R!. Unfortunately, neither underpromotion is strictly necessary, as White could instead win more prosaically by just releasing the stalemate trap with, say,
86.g5+ followed by promotion to queen. But still, the rook promotion is the move Sajtar played, and it's also the best, yielding mate in three.
The next position is from the game Tomic-Winzbeck (Dortmund Open 1993):
[FEN "2rRb2k/2P4p/6pP/p2B1pP1/Pp2pP2/1P2K3/8/8 b - - 0 1"]
1...Rxd8 2.cxd8=B! (2.cxd8=Q?? $10) (2.cxd8=R?? $10) (2.cxd8=N $18) 2...Bf7 3.Bf6+ (3.Bxf7?? $10) 3...Kg8 4.Bxf7+ Kxf7 5.Bd8 $18
43...Rxd8, to which White responded
44.cxd8=B!, since promotion to either queen or rook would yield stalemate. Not unlike before, while promotion to anything stronger would give up a draw, it's also the case that the bishop promotion wasn't the only option: a knight promotion would win easily enough here too. Still, the bishop promotion was actually played, and is the most natural, immediately threatening
45.Bf6#. Black resigned after the promotion, but could have set up one more not-so-subtle trap by playing
45.Bxf7?? would again be stalemate. White would instead need to play something like
45.Bf6+ Kg8 46.Bxg8+ Kxg8 47.Bd8 and win easily with the extra bishop.
This is a bit off-topic, but since we're talking bishop promotions, I can't resist mentioning another example from "actual play," this one the rather silly online blitz game Rybka-Nakamura (ICC 2008). The six bishop promotions in this game are of course entirely unnecessary, but the game itself is a good illustration of how an engine can go horribly wrong in a completely closed position in which it insists on trying to win with an entirely illusory material advantage that it has. The final position after
[FEN "K7/2bb4/1bbb4/3k4/8/8/8/8 w - - 0 1"]