There is no methodology here as to whether the test was local (~0 latency) or simulated-real (50-200ms roundtrip latency). The latter amendment might change the picture substantially.
Hi Jeremy, I'm the author of the post. The tests were not conducted locally, but over a real network, albeit a low-latency one. Latency is definitely a key factor, so I'm planning to run some more tests under different network conditions. Thanks for your feedback.