As you found out the orchestrator itself already utilizes a considerable amount of resources. Therefore I don't see a real benefit in a single-host scenario (apart from the Kubernetes' unified API to operate containers or for development).
Though, especially when you have additional/multiple hosts (multiple Raspberry Pis or e.g. include instances of a public cloud provider to your cluster) you can leverage Kubernetes' scheduling capability: using e.g. DaemonSet you can ensure that "every node of the cluster" runs an instance of your workload.
Additionally, you could even limit the resource consumption of those workloads to leave some space for other workloads (or define priorities).
IMO: if you want to run Folding@Home in a container I'd rather go for something like Podman. I assume there is no interoperability between the Folding@Home instances, so they can operate independently from each other. No real need for an orchestrator.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…