"a less frequent term might be more important (to receive attention in an ESL class) if it is a superordinate of other more frequent terms whose similarity/semantic relations might not be obvious...My feeling is that although his idea is intuitively appealing, it depends on choosing the right examples. As a general rule, hypernyms will be more common than their hyponyms (up to the point of abstraction*). Fruit occurs 40 times per million words in the BNC, while apple occurs only 26. Cooking apple turns up but once, and newtown wonder (a kind of cooking apple) occurs not at all. If we take this the other way, food comes in at 187. This certainly looks like a good rule of thumb.
The point is that although "fruit" might be less frequent than "apple" or "pear", it could be considered more "important" because it acts as a vessel into which other instances can be organized and remembered. And conversely, "apple" may be more frequent than "pear" but equal in its "importance" for vocabulary learning."
But it isn't the only possible relationship. If we look at rhetorical tropes, we get a very different pattern. Metaphor scores 9 to figure(s) of speech's 0.5 and trope's 0.18. If we drill down, we get mixed metaphor, dead metaphor, frozen metaphor and synesthetic metaphor, all of which are necessarily less common than metaphor itself.
I think it would be a tough job to convince anyone that trope is a more important word to teach than metaphor merely by dint of its being a direct hypernym. Nor would you get many takers for zeugma or synedoche just because they're sister terms. I don't think it is semantic relationships that matter but frequency.
*When I say "up to the point of abstraction", I mean that once you get higher than 'food' you start dealing with more abstract concepts like substance, matter, or physical entity, none of which are likely to be particularly common.