obviously the connector cannot handle aggregations. Direct Discovery is a feature which uses SQL aggregation queries only which consists of a GROUP BY clause and aggregation functions (count/min/max/sum/avg etc. for charts).
Usually this functionality is only provided by a real database. I wonder if you're going to implement this in a connector..
You could place another database (maybe an in-memory database) between your source database and QlikView which gets loaded once a day or so to serve the Direct Discovery feature independently from the original source.
Direct Discovery is not currently supported within QVX, only for the ODBC based connections. As for Qlikview soliciting the source db, there are some parameters you can set to throttle and control the load that is placed and also the duration for refresh of a Qlikview side cache that will hold previous results.