Is this perhaps another instance of #3851? http://hackage.haskell.org/trac/ghc/ticket/3851

Honestly: I don't know.

My example is different from the one shown in #3851 in that it also does not work in GHC 6.10 (which even panics instead of giving the error 6.12 gives) and in that it uses a data family, not a type family.

Whether or not it is the same "awkward interaction" that causes this behaviour is beyond my expertise.

Cheers,
Sebastian

-- 
Underestimating the novelty of the future is a time-honored tradition.
(D.G.)