I must not be very clever as it took me 3-4 attempts to understand the before last paragraph (e.g. that without Feather it took 20 minutes while with Feather only 2 minutes to load a video).
Be sure to read it a couple times if you run into the same initial impression.
The clear point being made is that in many parts of the world, the original heavy page was taking a long time to load. Videos would only start after 20 minutes. Feather dramatically reduced the initial load time, but because of what was being measured (video start times), it looked as if Feather had increased load time. In reality, videos were being watched 18 minutes sooner.