As noted in https://support.atlassian.com/browse/JSP-27005, the single biggest performance issue when scaling Jira to a large number of issues comes from inefficient date sorting performance in Lucene.

The strategy to solving this is essentially the same strategy described in JRA-15549. Upon any change to an issue, a new Lucene IndexReader/Searcher object should be generated in the background and "pre-warmed" by doing any search that triggers a sort on a date field. Once the object is ready, the active object should be swapped out for the newly generated one.

There is a trade-off here in that search results could potentially be a few seconds out-of-date, but interactive performance of searches that sort on dates should be greatly improved.

It is possible that JRA-15517 might make this feature unnecessary, but I won't be able to report on that until Jira 4.0 is released.