My understanding is that prepacking exists to make the structure more compatible with Rosetta’s scorefunction. If there are bad rotamers or rotamers with clashes, docking’s Monte Carlo protocol will be biased towards trajectories that allow the bad rotamers/clashes to be at the interface (so they can be fixed) regardless of the actual best docking orientation. The energy improvement from fixing a bad rotamer will outweigh the improvements from a better docking orientation, leading to wrong docked structures. Basically, prepacking smooths the energy landscape for docking and removes false minima that exist due to quirks of the input structure.
.
