<bXY> is parsed as <b> if XY is a non-ascii character.
Examples (also included in the URL):
<b→> doesn't work! </b>
<bä> doesn't work! </b>
<boo> works fine </b>
The parser misinterprets
and originally seen at https://en.wikipedia.org/wiki/Special:Version/Credits/BetaFeatures where Siebrand's email-address is causing the issue.
@Legoktm noted: To reproduce you need to have $wgUseTidy = true;, but the parser doesn't actually have to call MWTidy::tidy()