Old rankings should never be changed unless we really really have to.
We already have a logical reasonable vs. extreme division (or just single ranking where applicable) in all of them. Dividing by cpu sockets would only make some marginal differences in pcmark05 and 3dmark06 iirc (in the latter only the bot AIs in cpu tests get multithreaded extensively and this workload seems to be quite limited), so there was/is no need to do it.
The idea would be to never have more than two rankings for a new benchmark, because more than that leads to a confusing situation, and can give too much emphasis for a single bm. So the division (or opting for just a single category) would be made depending on the particular (new) benchmark, looking at how it scales. We could divide by cpu sockets, or by gpus, or by both of them, but would always have at most two rankings for any benchmark.