Thanks a lot, Stalker!
In addition i've discovered that function file_get_extension in core\file_api.php does not work correctly with unicode (e.g. Russian) file names because of using string variable as an array of chars.
To fix this you have to replace "$p_filename[$i]" by " mb_substr($p_filename,$i,1)" (keep in mind that mbstring library should be enabled in your php.ini).

Just replace whole function file_get_extension in core\file_api.php with following:

--------------------
# Get extension given the filename or its full path.
function file_get_extension( $p_filename ) {
    $ext        = '';
    $dot_found  = false;
    $i          = mb_strlen( $p_filename ) - 1;
    while ( $i >= 0 ) {
        if ( '.' == mb_substr($p_filename,$i,1) ) {
            $dot_found = true;
            break;
        }

        # foung a directoryarker before a period.
        if ( (  mb_substr($p_filename,$i,1) == "/" ) || (  mb_substr($p_filename,$i,1) == "\\" ) ) {
            return '';
        }

        $ext =  mb_substr($p_filename,$i,1) . $ext;
        $i--;
    }

    if ( $dot_found ) {
        return $ext;
    } else {
        return '';
    }
}

One more remark about encodings.
We know that some attached files could be displayed in bug page in [show content/hide content] block (e.g. php, txt and so on).
When such file is attached to issue it's content will be downloaded and written inside span class ant then it will be shown in page.
The problem is that most text files in russia saved in windows-1251 encoding and mantis now works in utf8. In this case text in cp1251 shown in utf8 page without any conversion.
To correct this, go to core\file_api.php an look throu file_list_attachments function.
Find string

echo htmlspecialchars( $v_content);

if (mb_check_encoding($v_content,"cp1251")){

$v_content=mb_convert_encoding($v_content,"UTF-8","cp1251");}

Thank you very much.

I believe mantis code should be rewrited so that every php non-utf-aware function is eliminated (or its arguments is preprocessed if possible) or replaced with mb_* equivalent.

But mbstring is not enabled on most non-russian configurations. That is why using mb_* stuff should be conditional, if mbstring lib is available.

IMHO the best solution is to implement a layer with equivalents to all non-utf-aware php stuff, which internally checks mbstring presence.

I think it's fixed in 1.3.