There seems to be a bug in the localization for strnatcmp and strnatcasecmp. I searched the reported bugs and found a few entries which were up to four years old (but the problem still exists when using swedish characters).
These functions might work instead.
<?php
function _strnatcasecmp($left, $right) {
  return _strnatcmp(strtolower($left), strtolower($right));
}
function _strnatcmp($left, $right) {
  while((strlen($left) > 0) && (strlen($right) > 0)) {
    if(preg_match('/^([^0-9]*)([0-9].*)$/Us', $left, $lMatch)) {
      $lTest = $lMatch[1];
      $left = $lMatch[2];
    } else {
      $lTest = $left;
      $left = '';
    }
    if(preg_match('/^([^0-9]*)([0-9].*)$/Us', $right, $rMatch)) {
      $rTest = $rMatch[1];
      $right = $rMatch[2];
    } else {
      $rTest = $right;
      $right = '';
    }
    $test = strcmp($lTest, $rTest);
    if($test != 0) {
      return $test;
    }
    if(preg_match('/^([0-9]+)([^0-9].*)?$/Us', $left, $lMatch)) {
      $lTest = intval($lMatch[1]);
      $left = $lMatch[2];
    } else {
      $lTest = 0;
    }
    if(preg_match('/^([0-9]+)([^0-9].*)?$/Us', $right, $rMatch)) {
      $rTest = intval($rMatch[1]);
      $right = $rMatch[2];
    } else {
      $rTest = 0;
    }
    $test = $lTest - $rTest;
    if($test != 0) {
      return $test;
    }
  }
  return strcmp($left, $right);
}
?>
The code is not optimized. It was just made to solve my problem.