Well, the floating precision issue can be worked around by explicitly calling mkDouble with the number as a string, ie:
[ (mkTuple [ (mkDouble "-0.65740735740229495") (mkDouble "2.5278185274873568") ]) ]
[ (mkTuple [ (mkDouble "-0.6600253512802865") (mkDouble "2.5301456447922108") ]) ]
which at least avoids having the dconf values change back and forth all the time.
Hmm… nixpkgs-unstable has 0.0.12; seems like that should be bumped.