This might be traceable to custom probes or some configuration change. I would go to the history actions panel and see if the change is still there that was done the day the dude started consuming memory. Disable the probe or remove the change and see if things improve.
If you can’t correlate it to a change maybe find/run sqlite tools against the database and do an integrity check (I don’t know how) and a vacuum (I Documented it here somewhere…).