It sounds like they mean that, if you have all the data you want for your skiplist and all you need going forward is a read-only data structure for doing lookups, then you can put your data in order in a regular list and then build up the towers of skip links in deterministic fashion to get an optimally balanced distribution.

Doesn't seem like much of a trick, though. If you can do that, then you can just stick it all in an array and bsearch to find things.