I get the feeling the biggest problem with site-local search engines is a tacit requirement that the search index always must be up to date. That severely hamstrings any search engine, since there's a wealth of supplemental information to be gathered by considering the corpus as a whole that is simply not available if you support real-time updates.
It's made worse by the fact that it's usually the only way of navigating confluence, as any non-trivial confluence eventually turns into a nightmare maze of abandoned stale pages, dead links and half-baked attempts at restructuring it where the person enthusiastically pushing for the restructuring effort sort of gave up a third through because it turned out to be a lot more work than it seemed.
I've seen this time and time again in both big and small organizations that use confluence. Makes me feel there is a fundamental design problem with the product.