To explain further what I mean, this testcase currently throws error: XML5111: WC_E_XMLCHARACTER: illegal xml character in IE9.
With responseText already being decoded to a native JS string, you can not really properly interpret responseText as another encoding anymore.
Although actually the latest versions of the other browsers actually seem to have worked their way around it, looks like they are not just doing encoding sniffing, but even content sniffing as when I enter ISO-8859-5 in the encoding (test file 3), responseText will show a different character.