We looked at Sphinx but ultimately decided to use Solr/Lucene instead. There were a couple reasons:
* We're not doing fulltext search, but searching by a set of tags that we didn't want tokenized/stemmed. Doing this with Lucene was easy, while trying to get Sphinx to work this way really seemed like it was going against the grain.
* The incremental update support seemed a whole lot simpler in Lucene than it was in Sphinx. The exact same stuff might be going on under the hood, but adding new documents seems to "just work" in Solr with less worrying about the implementation details.
From everything I've seen, Sphinx's performance is excellent and it's very well-suited to fulltext search. It just seems a little less flexible than Lucene.