Interesting! I may not have found it because I dislike Youtube for technical content. It's definitly an interesting benchmark. Using replicas in Kubernetes, but not `cluster.fork()` feels like a complicated choice to me, but I don't know k8s that well.
However, it seems to me the results are broadly similar. The first test he describes as a "synthetic benchmark", and Bun performs better. The second test, "a real-world use case", has Node and Bun trading blows, with a slight advantage to Node. Happy to hear your interpretation!