And that benchmark, quite pitifully, looks like this:

Benchmark: timing 1000000 iterations of Eval, S_op...
      Eval: 140 wallclock secs (133.80 usr +  0.03 sys = 133.83 CPU) @
+ 7472.17/s (n=1000000)
      S_op:  1 wallclock secs ( 0.81 usr +  0.00 sys =  0.81 CPU) @ 12
+34567.90/s (n=1000000)
[download]

with:

#!/usr/bin/perl -w
use strict;
$|++;

use Benchmark;

my $string  = 'cat';
my $search  = 'c';
my $replace = 'b';

timethese( 1000000, {
                  Eval => sub { eval "$string =~ tr/$search/$replace/"
+ },
                  S_op => sub { $string =~ s/$search/$replace/g },
                 }
          );
[download]

Man, that's ugly.

--
Casey