The search you start with the match op is completed with the substitution op because they both use the same iterator (pos($file)). As a result, your match op always starts from the start of $file.

my %translations;
my %scrubbed;
while ($file =~ /^ \s+ description \s (\w\w\w\w(-\w+-\w+)) ;/xmg) {
   my $real_dev = $1;
   my $name     = $2;

   if (!$translation{$real_dev}) {
      # Make sure not to duplicate anything in %scrubbed.
      my $scrubbed_dev = ...;

      ++$scrubbed{$scrubbed_dev};
      $translation{$real_dev} = $scrubbed_dev;
   }
}

my $pat = join '|', map quotemeta, keys %translations;
my $re = qr/$pat/;

$file =~ s/($re)/$translations{$1}/g;
[download]

If you just want to change the device description instead of every occurrence, I suggest

my $output = '';
my %translations;
my %scrubbed;
for my $line (split(//m, $file, -1)) {
   if (
      my ($pre, $real_dev, $name, $post) =
         $line =~ /^ ( \s+ description \s ) (\w\w\w\w(-\w+-\w+)) ( ; .
+* ) /xs) {
   ) {
      my $scrubbed_dev = $translation{$real_dev};
      if (!$scrubbed_dev) {
         # Make sure not to duplicate anything in %scrubbed.
         $scrubbed_dev = ...;

         ++$scrubbed{$scrubbed_dev};
         $translation{$real_dev} = $scrubbed_dev;
      }

      $output .= $pre . $scrubbed_dev . $post;
   } else {
      $output .= $line;
   }
}
[download]