Nevermind, I mixed something up. Anyway, with pooled sequences i.e. when using the sequence generator with a proper cache size you would have better performance and it would be possible to rule out possible PostgreSQL specifics, but I don't think this would make a difference here.
You could try to debug to org.hibernate.id.insert.AbstractSelectingDelegate#performInsert and find out why you have a different id value returned with the newer Hibernate version.