I don't think that it was working because there were some internal caches instead, it was part of an internal agreement between Docker and GCP, as reported here: https://www.googlecloudcommunity.com/gc/Google-Kubernetes-Engine-GKE/Why-did-I-get-an-email-about-Google-IP-addresses-subject-to/m-p/739854

We are now having several issues with GKE clusters and docker hub pull rate limit, when the token is not configured, even before 15 July everything worked as expected.