My opinion is that there isn't anybody else providing the numbers because it's impossible to get a correct number. Suppose Google homepage runs PHP, google.com/a uses Go, google.com/b uses Django, netflix.com use a custom framework running on Node.js, and facebook.com uses an unknown framework. W3 will happily tell you that the entire google.com is powered by php, fail to attribute netflix to JavaScript, and won't count facebook.com at all. That's exactly what's happening with their methodology. They only attribute each website at most once and rely on decade old hints from headers and error pages that are often non existent on newer/in-house web frameworks these days.

Want a real example? Everybody knows that Instagram runs Python. But https://w3techs.com/sites/info/instagram.com can't tell which server side language it runs. There you have it.

I would not use the number unless it can be validated, rather than use it simply because it's "the best we have". No, it's not even remotely good.