Great posts, Nietzsche.

Scott, chesslab is notoriously inaccurate for these stat exercises. I don't know how they get their games, but it's a very random selection.

The Tartakover is out of fashion partly because 5 Bg5 is out of fashion. 5 Bg5 is out of fashion at least in part because the Tartakover is so good.

"I fear the QGD may be soon relegated to an ancient opening that modern GMs and computers have busted."

Not a chance. The QGD isn't going anywhere. 
Neither is the Ruy Lopez, the Sicilan or the Nimzo.
But let's talk about the numbers first.

According to Megabase (and TWIC), black has scored 45% in the Alatortsev so far in 2008.  The black avg rating was 2369 with an elo performance of 2382.  In other words, black is doing more than fine with 3...Be7 so far this year.  After the mainline 4.cxd5 exd5 5.Bf4 c6 6.e3 Bf5 7.g4!? Be6!  Black is better than 50% this year and is exactly 50% for the past 3 years.  I think the Alatortsev is fine.  In fact, it is statistically very healthy.

As for 5.Bf4, let's not forget that many white players take this option primarily against lower-rated opponents (to avoid the solid Tartakower and the drawish Lasker).  For example, so far in 2008 the average elo difference between White and black is more than 50 points.  In 2007, it was nearly an 80 point difference.  This helps explain why the percentages are a bit more than the usual 54% or so. 

One powerful example is in the line 5.Bf4 O-O 6.e3 Nbd7!?  (an interesting alternative to ...c5 used by Spassky, Karpov and Kramnik).
White has scored exactly 60% the past 3 years in this line. That sounds great until you see that his elo performance is actually just 10 points higher than expected.  For white, this is actually a rather low number since you usually want a +20-25 performance with the white pieces (roughly).  There are lines where white has more than 55% but is actually underperforming his elo.  There is more to consider than just the percentage.

For the past 3 years, in the Bf4 mainlines:
8.cxd5 scores 67%, but drops to 62% when black plays 8...Nxd5.
But white's had an over 80 point elo edge!  In fact, after the mainline 12...d4, white is down to 57% already and only scoring an average elo performance with the white pieces.  The interesting thing is that no black player within even 100 elo points of his opponent has lost from this position in over 3 years.  Its really very drawish.

8.a3 scores 60% (59% after 8...Nc6) but white has a 2464 performance with an average rating of 2477.  Basically, white is doing -13 elo in this line.  Not good.

8.Qc2 scores 62%, but is only +16 elo.  however, this might be the way to play it since it's a bit less drawish.

Anyway, the QGD is not busted.
Major improves can (and are) found in such lines.
Also, we see Kramnik, Anand, Topalov, and all the elite players still taking the black side in these openings in the past few years. Kasparov and Karpov played both sides quite a lot.  Basically every World Champion played the QGD for a reason.

Anyway, there is a lot more to openings then statistics, since numbers can be interpreted many ways.  Let's just assume that the Queens Gambit Declined is not going to be refuted anytime soon. It really is one of the great, classic defenses.

Cheers,
Nietzsche

ps - as a previous post mentioned, I think the reason for a lack of Tartakowers is that white is going into the exchange lines, Bf4 lines, and the Catalan.  Black is doing very well in all the mainlines of the Tartakower (over 50% in fact).  There are very few defense to 1.d4 with its pedigree and statistical success.  Hence, a lot strong whites are avoiding these lines against equal or stronger opponents.

Yes, I think I read somewhere that 5.Bf4 has a very good score for White in recent years (sorry don't have a database to back this up) so even if Black tries to duck the 'more dangerous' Exchange variation by using the Alatortsev (3...Be7) he still has to be prepared for Bf4 and no Tartakower....

I really don't think the Tartakower is going to be the next fad any time soon.  the positions have been heavily researched and black has found sufficient drawing lines.  MNb's stats of 52% score for White is actually below average. 

Taking into account the high number of draws compared to other openings, I just don't see this becoming a popular line at the highest levels.  It's still perfectly playable for those of us who aren't in the super elite tho!

I think the reason we don't see many tartakowers these days is that the entire chess world is occupied playing sharp semi-slavs and anti-moscow gambits. But the tides of fashion are continuously turning. Who knows in five years from now the Tartakower may be all the rage again.

I remember that Karpov, Nigel Short even Kasparov and Fischer use to play the Tartakower and they got excellent results with it. What happened after that? Isn't the Tartakower good anymore? Was it refuted? Doesn't it give enough winning chances? Was it left behind to avoid the exchance variation of the Queen's gambit? What is your opinion?