(when db.dispose is called in unitofwork test with sqlite, the first test that runs in memusage grows by two gc'ed objects on every iteration; then the problem vanishes. doesnt matter what test runs in memusage. doing a dispose() in memusage solves the problem also. screwing wiht the mechanics of engine.dispose() only fix it when both the pool.dispose() *and* the pool.ressurect() are disabled. its just a subtle python/pysqlite bug afaict)