The Dutch. I have played two games with now. In the first, my computer told me I had a disadvantage of 0.5; I won. In the other - I am sure HgMan will remember that one - according to Silicon Power I had equalized, while I realized, I faced serious difficulties. HgMan won.
Overall computers are highly unreliable in positions with unequal material, where tactics are beyond their horizon. So I am not surprised, Inn2 mentions the Slav-Botvinnik. It is probably also true for the Meran (Reynolds Attack and so on).

In another thread on computers and positional chess, I expressed my surprise at how computers had some difficulty assessing IQP positions when the rooks came off the board.  Inspired by MNb's comment that he might have to reconsider his correspondence repertoire in lieu of this, I thought it might be worth discussing lines that computers do not handle well, lines in which organic grey matter (research and preparation) trumped silicon speed.

I have found that computers typically underestimate the Bird (1.f4) as a bad move and it often takes a dozen moves for it to concede that White might have an advantage.  But surely there are some openings--especially of the closed variety--that correspondence players might find that they can use to good effect.  Inn2 mentioned computers weren't much help in the Semi-Slav Moscow/Botvinnik.  I presume the Dragon might fit this bill, too?  Others?