Caesar, there's nothing dubious about 3. ... Nf6. It's a perfectly respectable move leading to either the Classical (4. Bg5) or Steinitz (4. e5) variations of the French. I play it all the time as black.
The Steinitz is the far more common response by white, by a very large margin. Very few people actually play 4. Bg5 against me. I wish they would, so I could play the MacCutcheon (4. Bg5 Bb4) variation more often. I like it, as it gives sharp play along the lines of the Winawer, but white players are much less likely to know what they're doing against it.
The mistake in the variation given in the starting post has more to do with black castling. In the French, black should rarely castle, because white has a space advantage on the king side, and the potential bishop sac if he can get his bishop to d3. In many lines, the closed pawn center makes it safe for the black king to just stay in the center. There are even lines where black castles queen side, despite having played the c5 pawn break.
Looks more like a trap in the French Defense, rather than a main line. The line isn't listed in Chessbase Encyclopedia of Chess Openings 2011 as an opening. I think most players over 1500 would consider 3 .. Nf6 an inferior move. The fact that it does occasionally see high level play means it's playable, but relegated as a dubious line with potential.
The Chessbase Mega 2012 does have a bunch of games with that particular sequence of moves, but no high level player is going to push their pawn from c5 to c4 and fall for an obvious trap. The highest elo listed for black pushes a pawn from f7 to f6 in response. The bishop sacrifice falls apart after black pushes on move 10, the pawn from f7 to f6, because the bishop sacrifice only works if the king has no escape squares. I had a recent game that was similar, although actually moved the pawn a couple moves later, because I knew I had time to get my king an escape square.