TTree Draw gives different normalization

Hello,
I’m doing a very basic TTree::Draw and the normalization of the resulting histogram depends on which variable I’m plotting even though the sum of weights should be exactly the same.
Am I missing something obvious here?

Regards
Javier

lep_eta is float, jet_n is int, xsec is the cross section weight for each event.
The file I’m running over can be found here /afs/cern.ch/user/j/jmontejo/public/Powheg.root
The same result is obtained with versions:
5.30.01_slc5_gcc4.1_x86-64
5.34.07-x86_64-slc5-gcc4.3

root [1] gROOT->SetBatch(1)
root [2] TH1::SetDefaultSumw2(1)
root [3] particle_jets->Draw("lep_eta","xsec")
Info in <TCanvas::MakeDefCanvas>:  created default TCanvas with name c1
(Long64_t)14996424
root [4] htemp->Integral()
(const Double_t)1.36585610642543543e+02
root [5] particle_jets->Draw("jet_n","xsec")
(Long64_t)14996424
root [6] htemp->Integral()
(const Double_t)1.34125713252300557e+02
root [7] particle_jets->Draw("1","xsec")
(Long64_t)14996424
root [8] htemp->Integral()
(const Double_t)1.20497566223144531e+02

Adding a bit more information, if a do the respective htemp->Dump(), I can see for both of them:
fTsumw 137.325 Total Sum of weights

What is the difference in the computation of Integral() vs fTsumw?

Regards
Javier