Grumble... grumble... I hate doing speed optimization...
Golf is at least more logical, with speed opts strange things make a difference.

{
  my @rtoa;
  @rtoa[unpack  'c*', 'MDCLXVImdclxvi'] = (qw(1000 500 100 50 10 5 1))
+ x 2;
  sub roman_to_dec {
    my $n = 0;
    $n += $_ - $n % $_ * 2 for @rtoa[ unpack 'c*', shift ];
    $n;
  }
}
[download]

Faster than the reduce version, and faster than the uc(shift) version...