Running pods without CPU limits can lead to system starvation

Running pods without CPU limits: in our clusters we see a low usage of CPU resources and still our clusters need to run with a fairly large number of nodes. Ideally we’d like to decrease the CPU requests on pods, and allow the unused CPU resources to be used by any of the pods to avoid them being throttled even when the CPU is pretty much idling.
I found multiple articles arguing, that removing CPU limits is a viable solution:
https://home.robusta.dev/blog/stop-using-cpu-limits
https://erickhun.com/posts/kubernetes-faster-services-no-cpu-limits/

But the later is also stating, that by doing so a resource hungry pod might starve system processes like kubelet and as such bring the worker down (and eventually the whole cluster if the issue cascades).

Can anybody confirm if kubelet and other host OS processes can starve in case there are pods running without CPU limits on them?

kubelet has flags to prevent resource starvation for system daemons

https://kubernetes.io/docs/tasks/administer-cluster/reserve-compute-resources/

Thanks for the reference! This seems to be quite an old feature. I wonder how people could end up in situations, when pods managed to starve kubelet (assuming the feature was actually used)