When two engines play it, it may well be that it ends in a draw. For two humans however, black has a huge practical advantage if black is familiar with it and white is not. Black sees 4 Ng5 more often than white encounters 4...Bc5. Example: Anand-Beliavsky, Linares 1991: 0-1 in 44.
IM Lasse Østebø Løvik, It can be dubious, but suppose that you play the 4. Ng5 line against 2 Knights, And opponent plays ...Bc5, probably he is deeply prepared to all lines and you probably aren't. It is a very high possibility for you to blunder ( even if you're an IM ). All lines which gives an advantage for white require very acurrate play.
And tpr is true, there are about 350 Traxlers in the chessgames.com DB. Black win %60.
Yep, Traxler is dubious, but in practical games (esp shorter time-control), it's not that easy to play for white when he doesn't have an idea what he should do against it..
#8, you could also counter that argument; if white plays 4. Ng5 he should have a line at hand to play against the Traxler. This is mostly 5. Bxf7+ because that has just easy play for white, whereas Black needs to work hard to find something close to equality.
I have a lot of experience in the Traxler, and although I want to believe it's playable, I really don't think it is. In practical games against not-so-strong opponents, sure. In games against strong players you should not try the Traxler, imo.