The issue here is that recent Intel CPUs have an heterogeneous typology, there are 2 kind of cores: performance and efficient. The efficient one use less power, so it’s good for battery but they are bad at crunching data, performance core are the opposite. In turns out that Xen is not aware of this, so it does not know how to handle it and spread CPU time across cores without caring about the typology at all, which leads to non optimal power usage and non optimal performance.
I made a performance benchmark a while ago with P cores only, E cores only and letting Xen handle. Hybrid CPU: benchmarking performance when pinning to specific cores