(ITS#7473) back-mdb is 50x slower than back-hdb on datasets with numerous subtrees