Same here. Non Kubernetes project originated control plane components start failing beyond a certain limit - your ingress controllers, service meshes etc. So I don't usually take node numbers from these benchmarks seriously for our kind of workloads. We run a bunch of sub-1k node clusters.