I know that I have covered the following concern most recently in "Public Service Announcement" and more comprehensively in "The Proper Way to Blunder A Piece" but this recent game moves me again to action.
ckfuria - DaimonnomiaD
standard, FICS, 2014
1.e4 e5 2.Bc4 Bc5 3.Nf3 Nc6 4.Bxf7+ Kxf7 5.Ng5+ Qxg5
According to The Database, White's resignation is not inappropriate: the first player has scored only 17% after 5...Qxg5.
The Database has 473 examples of 5.Ng5+, with White over all scoring 24%. (Yes, some defenders chose not to take the piece.)
The idea behind White's extravagant Knight sacrifice is that maybe Black will overlook the attack on his Queen after 6.d4.
The Database gives 9 examples of 5...Qxg5 6.d4 exd4? 7.Bxg5; 14 examples of 5...Qxg5 6.d4 Bxd4? 7.Bxg5; and 3 examples of 5...Qxg5 6.d4 Nxd4? 7.Bxg5.
That amounts to only 5% of the time that White "surprises" his opponent - hardly enough to justify the sacrifice of the piece.
After all, the Jerome Gambit's sacrifice of one or two pieces itself scores over 40%!