Description
I've been digging into why BlockTree's Terms.intersect impl is slower for a "very finite" union-of-terms test over random IDs (LUCENE-3893) and I found a few performance improvements that at least for that one use case gets a ~11% speedup.