There was a bug in Cirrus that Cirrus always used 4CPU config for Windows Containers even though everything was billed according to the resources configured (in your case 2 CPUs).