I double checked against my non-Lammps REBO code that I used back when I was a grad student with Brenner. I get -7.774 ev as well. Just because the Brenner paper fits to a value of -7.4 ev, doesn’t mean that Brenner actually got that number with the parameter set he chose. Knowing Don, he probably hardwired an LJ potential into his code to do multilayer graphite and would have gotten the correct answer. He just didn’t note it in the paper.
In short, I think the lammps version of rebo reproduces the publish version in this case. One would expect a slight difference between multilayer graphite and single layer graphite because of the longer ranged dispersion forces. AIREBO, which includes the LJ interactions between planes, corrects for this error, and as you noted, should give you the correct energy.
Hope that helps,