I can reproduce this on 3.0.14 as a module and as a standalone cgi. Just run for
extra entertainment. It ocassionally explodes my Netscape too - is apache not too graceful on handling SEGVs in modules?

<?

function decode_mime_string ($string) {
global $i;
   if (eregi("=?([A-Z,0-9,-]+)?([A-Z,0-9,-]+)?([A-Z,0-9,-,=,_]+)?=", $string)) {
      if (ereg("^=?", $string)) $str = ' ' . $string;
      $coded_strings = explode(' =?', $string);
      $counter = 1;
        echo "\n\nthis is the string --$string-- and this is $i";
        var_dump($coded_strings);
        echo "\n\n";
      $string = $coded_strings[0]; /* add non encoded text that is before the encoding */
      while ($counter < sizeof($coded_strings)) {
     $elements = explode('?', $coded_strings[$counter]); /* part 0 = charset */

     /* part 1 == encoding */
     /* part 2 == encoded part */
     /* part 3 == unencoded part beginning with a = */
     /* How can we use the charset information? */

     if (eregi("Q", $elements[1])) {
        $elements[2] = str_replace('_', ' ', $elements[2]);
        $elements[2] = eregi_replace("=([A-F,0-9]{2})", "%\\1", $elements[2]);
        $string .= urldecode($elements[2]);
     } else { /* we should check for B the only valid encoding other then Q
*/
        $elements[2] = str_replace('=', '', $elements[2]);
        if ($elements[2]) { $string .= base64_decode($elements[2]); }
     }

     if (isset($elements[3]) && $elements[3] != '') {
        $elements[3] = ereg_replace("^=", '', $elements[3]);
        $string .= $elements[3];
     }
     $counter++;
      }
   }
   return $string;
}

for ( $i = 1; $i < 35 ; $i++) {
decode_mime_string("[brain] =?iso-8859-1?B?RWR1Y2Hn428g4CBEaXN04m5jaWE=?=");
}
echo "Done.";
?>