Mime headers can be encoded with different charsets. mairix caches words ignoring this fact, For example cyrillic:
> From: =?windows-1251?B?0O7n4CDF8OPg6/vl4u3gINHg4+Dr4eDl4uA=?=
> Subject: =?KOI8-R?B?1yDQ0s/Ez8zWxc7JxSDMxcvDyck=?=
If mairix can't work with UTF charsets, it at least should honor current locale's charset and convert decoded words to it. I think it costs just a few line calling iconv.
The last phrase supposed to be:
> I think it costs just a few lines of code calling libiconv.