> I presume it's nice that you can actually customize xmonad with code. So an appropriate DSL would be a programming language so using Haskell makes sense. Any other DSL would be more limiting.

You've hit the nail on the head here --- the problem is that xmonad offers an unlimited scope for configuration, including extensions to the behavior of xmonad. I'd rather see a clear separation between mechanisms for configuration and extension, even if that reduces the power of configuration.