A random embedded image wouldn't be passed through the HTML parser (at least not if the web server has it's MIMEs set up correctly - and if they don't, then the site would be considerably less usable).
It will be interesting to see the results though - to see if the null characters even reach that far or if this is effectively hitting Opera with a zip-bomb
Well yeah, I nearly added that disclaimer myself but then thought it was pretty obvious the previous poster was referring to referenced URIs rather than data URIs. Though I may have misinterpreted his post.
It will be interesting to see the results though - to see if the null characters even reach that far or if this is effectively hitting Opera with a zip-bomb