From the summary it would look like an utterly stupid technique with O(n) retrieval cost. However, I suspect the linked list they refer to are linked lists off

From the summary it would look like an utterly stupid technique with O(n) retrieval cost. However, I suspect the linked list they refer to are linked lists off each hash bucket, which in the case of collision have to be searched anyway so purging stale data is virtually free. It doesn't cost that much to search the entire list rather than just up to the match because most hash tables are designed to keep these lists short enough to support the "constant time access" performance claim. That's clever enough if you don't mind that some stale data might not be purged for a long time because its bucket not hit. However, it's also the sort of thing that any experienced programmer might cook up with a half-hour's thought. Which is the basic reason software patents are inherently anticompetitive: to a patent office, there's no such thing as an obvious idea. Partly because none of this is obvious to lawyers, and partly because the patent system was designed for chemical formulas and paper clips in an industrial age where if an idea wasn't already patented it must be patentable.