I wouldn't be very excited about adding a 4B param model to my database deployment, but using this kind of approach while testing an app to identify query plans where Postgres is leaving performance on the table seems valuable without much risk.