I didn't see a previous post on this issue, even though it's fairly destructive.
When the program wide SJIS/JIS conversion is turned on, then this will occur *regardless* of whether a channel is instructed to display and encode UTF8. If both are used (conversion + utf8 de/encode) then things go horribly wrong, as SJIS character codes will get converted twice.
This means that for instance テスト (tesuto in japanese katakana) becomes the highly garbled utf string 繝・せ繝 instead.
(NB: ubb for some mad reason converts unicode to escaped html codes so they show up on the page as codes, rather than as source... I can't do much about that)