The importer brings in table cardinality, but for even better query optimization, we should also bring in distinct value count, or provide an option to do so.

This may be a larger one-time cost during import, but it's well worth it, and definitely is faster than doing it by hand for every column in every object in SFDC. It greatly impacts performance, especially for dependent join scenarios, since many SFDC columns/fields happen to have very low distinct value counts.