The performance impact usually is negligible. Containers are nothing else than cgroups and a set of namespaces. In fact, you can create a container without any container runtime (podman, docker etc.). It might be that the performance hit was due to an image being built poorly, or the runtime being configured in a strange way? The only metric where there is some performance hit is the network, and that’s because - depending on the configuration - the traffic **might ** flow through more hoops. Obviously it is possible to run the containers in the host network namespace, if this is really an issue.

All of this not to try to convince you or to claim your experience is false, is just that I am very surprised, I am aware that containers have some downsides, but usually performance is not really one of them.