After enabling the MultiSourceJobQueue in production, we've had a serious backlog on GCE builds: - [incident link](https://manage.statuspage.io/pages/pnpcptp8xh9k/incidents/4qj8t3220jy8) - [incident-response discussions](https://travisci.slack.com/archives/C0254QM3Y/p1505731108000167) This happened even without any http-jobs going through and was immediately resolved after changing the queue type from `amqp,http` back to `amqp`.