Deviations from symmetry in non-uniform refinement

We have a ~250 kD protein complex with D2 symmetry and after non-uniform refinement obtained a 2.9 Å map where the density for the central part of the tetramer has a local resolution around 2.5 Å. After building and refining one monomer, we rigid-body fitted copies of that monomer into the density to form a complete tetramer. To our surprise, we could see that the density for the four different copies was not identical. The density shows conformational differences of backbone and side chains of up to approximately 1 Å between the four monomers.
Can I draw the conclusion that the non-uniform refinement algorithm allows for differences between the asymmetric units although we run the refinement with symmetry? Can we believe this or is it an artifact?
Many thanks!

@marias. Maps produced by non-uniform refinement are expected to conform to the specified symmetry within the precision of the voxel grid. Had the map in question been postprocessed in any way (such as local filtering) before the monomers were fitted?
There may also be an appearance of asymmetry caused by inconsistent rigid-body fits of the monomers to their corresponding sections of the map.
Please let us know if the map from a NU refinement job with symmetry constraints is intrinsically asymmetric and none of the explanations above apply.