Take a look at the Performance Benchmarks mentioned in the docs. More particularly, the Performance Summary section.
You could also perform some tuning using Namespace Isolation via Sidecar resource that can minimize resource consumption across the mesh.