There does appear to be a quirk where $upper() will not correctly replace a lowercase letter with its uppercase equivalent. The example that jaytea gave was $chr(223) which is a German lowercase character (ß). However, this link explains what is going on, and why it probably shouldn't be considered an mIRC bug (but rather a limitation with Microsoft's Unicode routines):
http://blogs.msdn.com/b/michkap/archive/2005/04/10/406880.aspxAlso to clarify, mIRC's case routines do support Unicode already, it's just that there are quirks like this one. The reason the OP didn't want to use mIRC's routines was because it was inefficient at counting the number of uppercase/lowercase characters in a given string compared to regex.
