I imagine their scaling problem isn't messages/day, it's probably lots of concurrent, persistent connections. And I don't think a connection pooler would work with this job queue setup.