Lets state the obvious, this is an imperfect and evolving measure and the goal is to foster discussion and rivalry in making our pages better, faster, and lighter.
Developers, designers, and product need to talk more on how to achieve this. A 1,700 word article might weigh 10KB but by the time you load HTML, JS, CSS, images, 3rd-parties, and ads, it can range between 2MB to 8MB depending on the web site. Bear in mind, the first Harry Potter ebook is 1.1MB and that includes cover art.
This uses WebPageTest.org to measure web page load speed on articles from a variety of news organizations and publishers and highlight the top ten.
Benchmarking across different sites is hard. The aspirational goal is measuring perceptual load time for a user - when the reader 'thinks' the page has loaded. For now this provides benchmarks to monitor.
Score = Load Time (seconds) * Speed Index * Page Size (MB)
This will be updated soon to also factor in Visually Complete and exploring Time To Interactive (TTI).
Tests run twice daily. Results are averaged of last 3 days on a simulated LTE connection. There's probably a better way of doing it but there you go. Eventual goal is to test on actual mobile hardware.
Questions? Problems? You can find me on Twitter at @donohoe
All data, tests, and formulae, is publicly viewable in this Google Spreadsheet.