I would be surprised if that theory held true. I expect that both very abstract (fruit) and extremely specific (golden delicious) items would have a lower sitelink count than the "golden layer of most useful terms" (apple) in the hierarchy (I am reminded of the theory of word length and term frequency in linguistics).
But I would assume that indeed in the subclass hierarchy that Wikidata will eventually exhibit would have such a "golden layer" (and that these terms are not randomly distributed over the hierarchy).
Would be fun to examine :)
Cheers,
Denny