Fhars has the numbers involved, and you can search the mailing list on this. What I remember is, Ian Hixie was spec editor and was then working at Google. He would do bigtable queries over "all the HTML pages in the world" in some sense, and find out things like that so many million pages had IMAGE tags. So he would float to the list that it should be handled as a matter of robustness. And most of the time those queries would be inspired by someone's suggestion, although I think he had gone ahead and tabulated a frequency table. This way you could judge whether potential new element names would clash with mistakes authors were already making.