...
Now the same plot but showing the individual CDE Recon energy versus its MC energy. There is indeed a population of CDEs contributing in a strange but linear way. This can happen for more than 1 CDE per event.
To see if there is a particular set of CDEs that are the cause of this, this is a histogram with the number of occurencies where CDE #(96*twr+12*lyr+col) has E(recon)-E(MC)>1 GeV. All CDEs can be triggered by this effect, it seems to be only a function of MC energy.
Now cutting on those weird events it is visible that E(recon)-E(MC) = 4*E(MC) when it happens.