A sound opening, or just a successful one

I saw a story the other day that listed the "most successful" chess openings. As with most studies of this type, the answer depends on what data set you use. I am pretty sure that 1.e4 e5 2.Qh5 Nc6 3.Bc4 Nf6 4.Qxf7# is probably the winningest opening ever, but if you only looked at games where the moves were actually recorded, then the strike rate for 1.e4 e5 2.Qh5 falls away.
The study that I saw had the Queens Gambit at the top of the list (for White), but I'm assuming this covered all the games that start with 1.d4 d5 2.c4 and did not differentiate between Declined, Accepted, Slav etc But it was the second opening which kind of had me bemused. It was the venerable Blackmar-Diemar Gambit! This indicated to me that the number of games studied were not just confined to 2600+ v 2600+ clashes, but probably included a fair chunk of online chess. It also seemed to say that the strength of the players was not a consideration either.
But numbers are numbers and it makes me wonder. Would you play an unsound opening that you knew would collect a lot of victims in the small pool you swim in, or would you stick with sound openings knowing that you score better against stronger opponents, even if your overall score was lower?

(Before you all write in, yes, I am an example of the first choice *cough* Traxler *cough*)