differences in DY spectra between 2.3 and 2.4

Dear all,

when validating mg5_amcatnlo 2.4 vs 2.3 in CMS we found significant differences in several distributions, as for example this one[1]. The plot shows the transverse momentum of DY reconstructed from two leptons at with "ME level" status codes. The same trend is obtained after parton shower [2]. The process is pp > e+ e- with up tp 4 additional jets using MLM merging. Looking at the log file, I saw that the cross sections before matching is about 25% lower in 2.4 in each multiplicity bin except 0 parton. Applying MLM merging the 1 additonal parton cross section remains different by -25% while at 4 partons the merging removes a lot of the events and the cross section is larger for version 2.4.

Studying the effect a little more by using these cards/model in standalone mg5_amcatnlo (no interface to cms stuff)[3], I saw that the jet selection to calculate the scale for alphas weighting was changed in ./Template/LO/SubProcesses/reweight.f
from
       elseif(ijet.le.njetstore(iconfig).and. i.eq. iqjetstore(ijet,iconfig)) then
              goodjet(j)=.true.
to
       elseif(iqjets(i).gt.0) then
              goodjet(j)=.true.

Can somebody comment a little more on the modification and let me know if the size of the effect makes sense? It seems quite large to me although the merging changes the the final net effect.

Cheers and thanks for any help,
Alexander.

[1] http://www.desy.de/~agrohsje/hidden/debug_mlm_mg2p2vs2p4/MCParticleAnalysis/dy_me_pt.pdf
[2] http://www.desy.de/~agrohsje/hidden/debug_mlm_mg2p2vs2p4/MCParticleAnalysis/dy_ps_pt.pdf
[3] http://www.desy.de/~agrohsje/hidden/mg5_amcatnlo_2p2vs2p4/