Difference in neutron simulation between geant4.10.4.2 and geant4.10.5.1

Hi all,
I am trying to calculate the neutron detection efficiency for a setup with four he-3 proportional tube counters. The counters are submerged in water and can move around to set different source-detector distances. Water is used to thermalize neutrons emitted from the source to increase the interaction cross-section with he-3. I am using an AmBe neutron source with a known neutron emission rate. For the physics list I am using QGSP_BIC_HP and G4ThermalNeutrons in my main src file as follows:

  G4PhysListFactory physListFactory;
  const G4String plName="QGSP_BIC_HP";
  G4VModularPhysicsList *pList = physListFactory.GetReferencePhysList(plName);
  G4int verbose=1;
  G4ThermalNeutrons *tn = new G4ThermalNeutrons(verbose);
  G4cout << " Registered G4ThermalNeutrons" << G4endl;
  G4cout << " User initialization done for physics list " << G4endl;

My question is- is it the right way to implement the low-energy neutron simulation in geant4? If this is sufficient enough for an accurate simulation why am I seeing very different hit counts in the detector-sensitive area for different versions of geant4? In geant4.10.4.2 ~9% of primary neutrons register energy in the detector-sensitive volume whereas in geant4.10.5.1 only 3% do. I am using the JEFF3.3 library for the evaluated cross-sections. Does anybody have an idea where things are going wrong?