There have been changes in the configuration of the Geant4 EM standard option 4
physics list between version 9.6patch-2 and 10.7patch-1.
The most significant one is related to the model used to simulate e-/e+ multiple Coulomb scattering
(below 100 MeV kinetic energy): the semi-empirical Urban model is used in 9.6 while the fully theory based Goudsmit-Saunderson (GS) model is utilised since version 10.4.
This latter also contains several important higher order corrections when used in the EM standard option 4 physics list since Geant4 version 10.4. These, together with its so called “error-free” stepping (see more details here), provide a significantly more accurate EM shower simulation compared to the earlier Urban model.
The cited report contains simulation results obtained by using EGSnrc, Geant4, MCNPX and PENELOPE. We expect, that your results (obtained with Geant4 version 10.7) moved closer to those in the report obtained by EGSnrc (and PENELOPE) compared to the reported Geant4 results (using version 9.6). This is because the above-mentioned accurate GS e-/e+ multiple Coulomb scattering model (as used in the EM standard option 4 physics constructor since version 10.4) is rather similar to that used in EGSnrc (the same theoretical foundation including all the corrections and the stepping algorithm). This increased level of agreement, between the simulation results obtained by using the new Geant4 GS model with its option 4 configuration and those provided by EGSnrc, have already been confirmed by several applications.
Note, that while we expect changes in the simulation results between Geant4 versions 9.6p2 and 10.7p1 due to the above-mentioned improvements (among others, since e.g. the parametrisation of the photoelectric cross sections have also been updated), we cannot tell you for sure that what you see is due to and only due to these as we have not made the corresponding simulations with both versions. However, you might try to run your simulation also with Geant4 version 9.6p2 (available here) that should reproduce the results provided in the cited report.