During the throttling wars, I seem to remember hearing somewhere that ISPs recognize speedtest.net runs and optimize for them, while also throttling competing content providers like NF. That is why fast.com was built: if ISPs throttle Netflix, they'll also be throttling fast.com so it will be reflected there.
This is true and still is. Most larger ISP will have Speedtest servers in their infrastructure. I remember installing one of their servers in 2010 in the ISP I was working for at the time. The thing it doesn't mask is last mile type issues. So if the node you're on in an HFC is oversubscribed and you start complaining about it, and can show this during peak hours using their own Speedtest server, they'll likely shuffle RF or split the node to relieve the congestion. In that case it does have value. You can also select the server you want to test against so it's not hard to validate speeds and latency outside of the, potentially, local-only test.
I leveraged Speedtest's CLI in a little mashup of Docker, Python, Grafana And InfluxDB because I had found some inconsistencies with other CLI speed testing applications and needed something to log a remote internet connection for a few months to prove bad and repetitive specific time of day latency [0].