This sounds like an X configuration issue.

No it isn't because MATE can use Xinerama. It is related to my XMonad.

I understand the original message as saying that it recognized screens correctly before xmonad was moved to a sandbox.

No, I said that MATE was recognizing the screens and it does it also now. I never installed XMonad from the package manager or from my user "global" cabal because last time I did it I had a lot of problems. XMonad in a sandbox would be perfect...

This makes me wonder if the tests are being done with a different xmonad library and/or the sandbox is somehow not seeing a Xinerama-enabled X11 (this can happen if the original was installed globally via OS package but the sandboxed build is using a local X11 bindings package, and the Xinerama developer headers/libraries weren't installed, for example).

I don't think we have enough information to diagnose this.

X11 bindings, XMonad and XMonad-contrib are all installed in the sandbox with Xinerama support. The Xinerama lib they used is the one installed by my package manager. Do you think I could give you some informations to understand what's going on?