It seems very strange, but during docking DockingSlideIntoContact routine sometimes does not work: after it is applied, proteins may still be separated by, say, 50 angstroms. In this case, sometimes there's a message in the output that DockingSlideIntoContact failed.
Moreover, these ambiguous conformations have energy very similar to solutions where proteins are in close contact and have low rmsd-to-native.
I use slightly modified docking script from pyRosetta site; the main thing I changed is maximum values of translation (15 A) and rotation (180 deg) of initial perturbation.
What is the problem?