I don't mind the averaging, but I think the non-averaging addition method of RMSS/HARP is better, in that it makes the game more accessable to new players. The old salts tend to average by rote without worry, but it does make the game mechanisms a bit less friendly to new people.
Switching from adding three stats to divide by three, vs adding three stats, works as the stat modifiers are roughly 1/3 of the RM2 version. . . .
To carry the logic out properly, should have been:
Stat x 3 + Racial RR mod x 3
So that arcane could be
All three stats + all three racial mods
And that math/method would work out
The racial stat mods were reduced in proportion to the stat bonus table drop from RM2 to RMSS, while the RR bonuses were not.
The RR Bonuses are thus a bit out of scale/methodology with the rest, leading to a hard skew in the racial RR bonus' affect on probabilities for hybrid and arcane, while leaving them identical for one-realm.
Since the racial mods in RMFRP Core look to be direct ports from RM2 character law, I suspect that's an "Oops" of unintended consequences, perhaps even something overlooked. I wasn't there, I won't claim to have any direct knowledge, but it looks unintentional, since in almost every other transfer over of the logic they tried to stick close to the RM2 probabilities with just easier methods for RMSS.
Shrug, with systems this detailed, there were bound to be a few points where things shifted, intentionally or not.