After further reading, especially <http://people.apache.org/~hossman/apachecon2006us/faceted-searching-with-solr.pdf>
(Thanks Hoss)
Depending on update patterns and index sizes, you can probably get better efficiency with multiple indexes, but not really more functionality (in your case), right?
Maybe I'm approaching your point of view : "Loose Schema with Dynamic Fields", this is probably my solution. There's something strange to me to consider a lucene index as a blob, but if it works for bigger than me, I should follow. So, it means one fieldtype by analyzer, and the datamodel logic is only from the collection side. I think I got my idea for september, but I would be very glad if you have something to add.
-- Frédéric Glorieux École nationale des chartes direction des nouvelles technologies et de l'informatique