It appears that our jobs generally run on VMs with 2 vCPUs. Consequently running
with 8 jobs will completely oversubscribe the machine. I suspect this is the
cause of #14453. Let's bump this down to 3 for now. Ideally we would determine
this from the environment.