From what I understand, cpu steal time is the CPU time that the VM kernel requested, but was denied by the hypervisor because it was serving another VM.
From my experience, on a XenServer that has dozens of VMs running, from time to time the steal time increases in both the dom0 and in the VMs (more perceptible in a small set of VMs).
What's the impact on the overall system performance of a high CPU steal time on XenServer's dom0? Does the dom0 has more priority than other VMs?