The problem is caused by TinyMCE's original zh.xml file. The file is encoded in utf-8 exactly, but the first line of it says <?xml version="1.0" encoding="gbk"?> . I just changed the bold string from gbk to utf-8, then update_lang_files.php can convert it very well.
This is a special case happens on Chinese only. So I think it is no need to patch the converting script. I can handle it by myself.