Is that claim backed up with analysis? Last year I tried 8...c5 in a fast correspondence game and got no equality after 9.e3, but a fast draw anyway. Probably 9...cxd4 is as stupid as it looks, but I found nothing really convincing and Rybka recommended it. 

[Event ""]
[Site "gameknot.com]
[Date "2010.04.21"]
[Round "-"]
[White "he"]
[Black "me"]
[Result "1/2-1/2"]
[WhiteElo "1988"]
[BlackElo "2307"]

1. d4 e6 2. c4 f5 3. g3 Nf6 4. Bg2 c6 5. Nf3 d5 6. Qc2 Bd6 7. O-O O-O 8. Ne5 c5 
9. e3 cxd4 10. exd4 Nc6 11. Nxc6 bxc6 12. Nd2 Ne4 13. Nxe4 fxe4 14. Be3 Ba6 
15. b3 Rb8 16. f3 exf3 1/2-1/2

I'm still mulling what openings I'm going to play and was looking through Starting out: 1d4! to see what John Cox had to say about various defenses to 1. d4.

In the section on the Dutch Stonewall, after 1. d4 e6 2. c4 f5 3. g3 Nf6 4. Bg2 c6 5. Nf3 d5 6. O-O Bd6 7. Qc2 O-O 8. Ne5 the book basically says that 8 ... b6 is the most common move, but maybe not best since White's last two moves are designed to meet it, fair enough. He also mentions that Mikhail Ulibin tried 8 ... Qe7 in a game that went in White's favor.

I decided to give this position to Houdini 1.5 to see what it thought. It came up with 8 ... c5 which isn't in my limited database.

That surprised me, and I had a hard time doing anything against it. After the simple 9. cxd5 cxd4 10. Nf3 e5 I prefer Black. He's got a nice central pawn mass, if he can defend it, and the White pawn on d5 can't be long for this world.

Houdini rates the position after 1. d4 e6 2. c4 f5 3. g3 Nf6 4. Bg2 c6 5. Nf3 d5 6. O-O Bd6 7. Qc2 O-O 8. Ne5 c5 9. cxd5 cxd4 10. Nf3 e5 as either dead equal or a very slight advantage to Black.

I tried 9. Nf3 instead, but after 9 ... cxd4 10. cxd5 e5 we're back in the same position, and if 10. Nxd4 then 10 ... Nc6 11. Nb5 and the computer still sees a small advantage for Black.

What do you think, is 8 ... c5 an improvement for Black?

btw: I did look through the big Stonewall thread, but didn't see this discussed, so my apologies if this has all been hashed out already.