Open Lopez : 'Long Variation'

In Unbalanced Material, I questioned Kasparov's comment that 'I have the feeling that this variation is avoided by both sides nowadays'. He wrote this in a note on the position that I diagrammed in that post, so I'll repeat the same diagram here.

To find the current state of theory on this position, I consulted Chessbase.com's database at Chesslive.de. The advantage of this database is that it includes many correspondence games, and I suspected that the diagrammed position has been well explored by correspondence players. The best of them like to play sharp, unbalanced opening variations which OTB players often categorize as 'unclear' (Informant symbol '∞').

Radio Match 1945
Reshevsky, Samuel

Smyslov, Vasily
(After 22...Qd4-e5(xP))
[FEN "r5k1/2p3pp/p7/1p1pq3/8/5PpB/PP4P1/R1B2R1K w - - 0 23"]

I found 57 games on Chesslive that reached the diagrammed position. Of those games, 56 continued 23.Bd2, and of those, 36 continued 23...Qxb2 24.Bf4 d4. This is the variation recommended by both Smyslov and Kasparov as the alternative to Reshevsky's 24...c5. Of those 36 games, 26 were correspondence games, the most recent an IECG email game from 2002. Of these 26 games, the 12 played since 1985 have resulted in seven wins for Black and five draws. Several of the games were played in national correspondence championships, meaning the players were expert or master strength.

The most recent OTB game between GM-level players was Tiviakov - I.Sokolov, Groningen 1994, which also finished as a win for Black. I conclude that White has gone wrong somewhere in the moves leading to the diagrammed position. But where? Kasparov wrote that 15.Qxd4, instead of 15.cxd4, is the move played nowadays. Is 15.cxd4 a weak move?