Hmmm ... it seems to me that your benchmarks are actually indicating that tr/// is slower than s///. Not what I would have expected, and apparently no one else did either, since despite the evidence, everyone is claiming that this test shows that tr/// is faster.

Update: I just ran the benchmark on my ActiveState build and got similar results ... that, strangely, s appears to be nearly twice as fast as tr in this trivial example. Results showing only about 1 million tr's per second vs. almost 2 million s's per second:

                   Rate transliterate  substitution
transliterate 1011345/s            --          -49%
substitution  1969797/s           95%            --
[download]

Just for background, this is ActivePerl 5.8.0 build 806.

UPDATE Update: Thanks to davido for pointing us in the right direction on this in the CB before he posted his node.

use Benchmark qw(:all);

$toto = 'this+is+my+text+and+here+is+more+and+more+this+is+my+text+and
++here+is+more+and+more+this+is+my+text+and+here+is+more+and+more';
$count =-5;


$results = timethese($count, 
            {
                'transliterate' => sub { $toto = 'this+is+my+text+and+
+here+is+more+and+more+this+is+my+text+and+here+is+more+and+more+this+
+is+my+text+and+here+is+more+and+more'; $toto =~tr/+/ /; },
                'substitution' => sub { $toto = 'this+is+my+text+and+h
+ere+is+more+and+more+this+is+my+text+and+here+is+more+and+more+this+i
+s+my+text+and+here+is+more+and+more'; $toto =~s/\+/ /g; },
            },
            'none'
        );
        cmpthese( $results ) ;

exit 0;
[download]

Now we have the same string for the benchmark tests! And the results are as expected:

                  Rate  substitution transliterate
substitution   73807/s            --          -88%
transliterate 625256/s          747%            --
[download]

Boy, do I feel stupid for not seeing that!