Mandrake Linux 9.1, gcc 3.2.2, Intel Xeon PC, G4.6.0 (patched).
Test case:
exampleN01 with follownig changes:
copper cube 10cmX10cmX10cm as World volume, no other volumes.
All primaries start in the center.
Physics List - standard one:
__________________________________________________________________
PhysicsList::PhysicsList():  G4VModularPhysicsList()
{
  // default cut value  (1.0mm)
  // defaultCutValue = 1.0*mm;
  defaultCutValue = 0.7*mm;
  SetVerboseLevel(1);

  // General Physics
  RegisterPhysics( new GeneralPhysics("general") );

  // EM Physics
  RegisterPhysics( new EMPhysics("standard EM"));

  // Muon Physics
  RegisterPhysics(  new MuonPhysics("muon"));

   // Hadron Physics
  RegisterPhysics(  new HadronPhysicsLHEP_BERT("hadron"));

  // Ion Physics
  RegisterPhysics( new IonPhysics("ion"));

}
___________________________________________________________
First I ran 1e+6 of 1MeV gammas, then (in the same simulation) 1e+6 of 31.7
MeV protons 3 times (total 3e+6 of protons).
I did it for 2 types of hadronic physics LHEP_PRECO and LHEP_BERT.
The following table shows virtual memory usage obtained by "top" program.
                                       LHEP_PRECO    LHEP_BERT
at startup                                   8900k          8328k
1e+6 of 1 MeV gammas             8912k          8352k
1e+6 of 31.7 MeV protons         10748k         18380k
2e+6 of 31.7 MeV protons         10768k        28200k
3e+6 of 31.7 MeV protons         10768k        37996k

Obviously there is a memory leak in LHEP_BERT the "leakage speed" is about
10Mb/(1e+6 histories), it will give 1Gb of swap space for 1e+8 histories :(