While the AWS slaves aren't "slow", we should prefer to run on our in-house capacity first. This will give the idle-slave-watcher a chance to shut down the AWS instances if they're not needed.
We need to allow for all the EC2 slaves being on bm35, while the slaves like the hp's are on bm12/13/14/15/25. So some logic that picks up jobs on hp's immediately, but ignores jobs for a couple of minutes if only ec2 slaves are available on the master.
I think I'm working on this
see also bug 795649 for improvements to how we should pick recent slaves.
Not sure how much this is still needed. We have far fewer in-house slaves now.
This is as fixed as we want it to be. A lot has changed since it was filed. Eg, jacuzzis.