This would exclude the many who use it not just on websites, but for personal offline documents too. I think any analysis should be on implementations rather than across websites, because it’s difficult/near impossible to gauge popularity of offline usage. Nevertheless, there could still be meaning in analysis across websites as that might dictate/relate to how markdown is used offline.
Either way, removing underscores is likely not going to happen. Markdown usually wants to use what’s already being used, and underscores as a means of emphasis is still common. But I do understand that’s not exactly the point here.
John’s babelmark2 is probably related here.
See also: Tables vs fenced code blocks, when is something “common”?