The 6.1 docs mention max_concurrent_searches, but it's not in the 2.3 docs or in any version of the Java API that I can see.
We're looking at supporting potentially-very-large multisearches from users; it's OK for them to be slow, but we don't want them to kill performance or for normal (non-_msearch) searches, or have those searches rejected because of a full search queue. Do we have any options other than breaking into small batches on our side and living with the extra transport overhead?