Importance biasing change the results

Dear G4 team,
I have been using importance biasing in order to score energy deposition in the case of neutrons incident to the skin.
I score it for different particles such as e+, e-, photons, protons and heavy ions.
I have made some tests and have seen that results with and without biasing are not in agreement, whereas I multiply the score by the step weigth:
G4double edep = step->GetTotalEnergyDeposit() * step->GetPreStepPoint()->GetWeight();

For instance, here is the scores I got in GeV/g for incident neutrons at 50 keV.


We can see a strong difference in the e- energy deposition that cannot be explained by the statistical uncertainty. I have observed that: lower is the neutron energy, higher is the discrepancy on e- energy deposition. At higher energies of primary neutrons, results become in agreement (above 1 MeV).

Do you have any idea or suggestion for what is happening? I cannot understand these results.
Thanks in advance.