As a couple of other folks have noted, lots of ISPs prioritize traffic to (some) known speed test sites, so differing results from different speed tests -- and

As a couple of other folks have noted, lots of ISPs prioritize traffic to (some) known speed test sites, so differing results from different speed tests -- and results that differ from real world experience -- are pretty common. Even weirder, sometimes this traffic prioritization happens partly in the firmware of the modem the ISP provides. On one of our test network connections, we can reliably trigger a firmware issue in the router that causes very high packet loss. Then we can "fix" the issue by visiting www.speedtest.net from any computer on the LAN behind the modem. (My company makes in-browser video calling stuff, so we spend a lot of time worrying about network connectivity.) And speaking of packet loss, I'd love to see Netflix add a measurement for UDP packet loss at several throughput levels. There's no easy way for non-technical users to do a really good packet loss test. We often get customer support queries that say something like, "My video calls are terrible but it's not a network issue because I did a speed test and I'm getting 100Mbs." Which is a totally reasonable thing to say. But when we look at call data, the person is experiencing regular spikes of, say, 15% packet loss. That's not something that will effect web browsing noticeably, or even Netflix video very much, but it makes for really frustrating video calls.