Ok, providing atomicity is a must have, the query then purge approach might not work. I still vote then for FullTextSession#purge(org.apache.lucene.Query query) where we initially limit the query types we accept to something we can easily "serialize". This will work whether or not a full query serialization will be possible in the long run or not.

IMO, it would be beneficial to investigate the Lucene query serialization problem first.