Not wishing to undermine a good IPv6 argument, I do wonder if the latencies from CGN and CLAT are worse than the Kubernetes "traefik" induced intermediaries and address proxying.<p>If K8s had been released pure V6 I wouldn't be asking this because the rules would be implemented differently. That said, HAProxy and/or some kind of iBGP intermediate to shard the load and do failover implies work being done, albiet not every packet (once the state is done to get to which backend handles which frontend query)