kube-proxy operates at L3/L4 while service meshes generally operate at L7 so it can load balance on a per HTTP request basis. Particularly useful for long lived connections commonly used in gRPC and others.