We expect the stress-energy tensor to have zero divergence, because this is required for local conservation of energy-momentum, which has been verified to high precision in laboratory and solar system experiments. The standard review article is Will, "The Confrontation between General Relativity and Experiment." Will does explicitly discuss PPN parameters that would violate conservation of momentum: http://relativity.livingreviews.org/Articles/lrr-2006-3/articlesu5.html#x11-190003.2 [Broken] What remains unclear to me is: (1) Why does PPN appear to treat energy and momentum differently, when they're actually part of the same four-vector? Or is this difference just in Will's presentation? If we assume (a) no preferred frame and (b) conservation of energy in some frame, then it follows that we must have conservation of momentum. But since PPN explicitly allows preferred frames, the situation seems hazy to me. (2) How would one test whether dark energy gives a stress-energy tensor with zero divergence? We can't do it in the lab, because we have no access in the lab to any empirical tests of the properties of dark energy (e.g., its equation of state). So do cosmological observations tell us anything about this, or are we assuming that dark energy has zero divergence purely because the rest of physics works that way? (3) At scales where dark energy has a negligible effect, which is the better test of conservation of energy-momentum: laboratory experiments, or solar-system/space probe data?