For cluster-wide setting you will probably want to drop that significantly -- start lower, as in somewhere around 10MB and work up from there as necessary. For the queries you are finding slow (the reason for these emails) you can set work_mem specifically for the session.