I took a longitudinal view of counting unique pageids grouped by CDN and graphed it over time to see which CDNs are gaining ground in terms of serving the base HTML and surprisingly Cloudflare is up and growing (expected Akamai ) but then we also included more URLs and this might be a function of all the newly included URLs being on Cloudflare
However if I remove the top 3 from the list (Cloudflare, Akamai, Google) we see the following pattern:
Incapsula has the highest growth rate But then the above is simply counting unique(pageids).
Lets do the same analysis for average number of bytes downloaded and we see that Google takes the cake (presumably due to youtube or volume of js frameworks!):
Now that we broke the unique pages and byte volume by CDN its only natural to ask how is the performance per CDN breaks down. For this we can choose 75th percentile of speedIndex and see its evolution over time:
Sounds like Cloudflare and Google are the fastest for the content they serve
(let me know if you want metrics other p75 and I can generate them for you)
If you disagree with speedIndex and like PageSpeed score that breaks down in similar way:
But then you argue that those are new metrics and maybe you want the break down based on fully loaded time (flawed but okay to indulge) and you see this pattern
Suprise! Google is the worst based on this but Fastly is the best. So choosing the right metric is key here. To close the loop lets measure the TTFB (time to first byte) which is a function of the cdn's capacity, geographic reach and software techniques involved in getting the first bytes of HTTP response header and should be a true measure of a good CDN and you see this:
The above is expected TTFB with Google being at 524 ms and more than a second for China Net Center. If you prefer a time view here you go
This is already a long post so will stop here but its very interesting to look at httparchive trends over time rather than one specific dataset or one specific metric
FWIW, I would not read too much into speedindex, pagespeed, and fully loaded times. If I'm following correctly, you're classifying pages based on which CDN provider serves the base HTML.. but those same pages then fetch resources from plenty of other origins (and CDNs). So, it's somewhat misleading to link CDN X to any one of those metrics.