You have a valid point in that node i was kind of comparing apples to oranges (Of course that node was about uppercasing input which is a bit different). I must have been having brain dead kind of day. As for your number I changed my benchmark code to this for this specific case to this:

#!/usr/local/bin/perl -w
use strict;
use Benchmark;
my $count = 900000;

## Method number two
sub One {
   my $data='for-bar-baz';
   $data =~tr/-/_/;
}
## Method number Two
sub Two {
   my $data='for-bar-baz';
   $data =~s/-/_/g;
}
## We'll test each one, with simple labels
timethese (
  $count,
  {'Method One TR' => '&One',
   'Method Two s'=> '&Two'
   }
);
exit;
[download]

Which results in the following

Benchmark: timing 500000 iterations of Method One TR, Method Two s...
Method One TR:  2 wallclock secs ( 1.87 usr +  0.00 sys =  1.87 CPU) @
+ 267379.68/s (n=500000)
Method Two s:  5 wallclock secs ( 4.84 usr +  0.00 sys =  4.84 CPU) @ 
+103305.79/s (n=500000)
[download]

FYI- I got these numbers using Perl 5.6.0 on Win32 You are right that this is not as big of difference as 17:1 (Which when I first ran seemed odd to me...) But is still a decent gap.
I got the suggestion about not using S when TR will do from Effective Perl Programming (co-authored by our own merlyn). It is a great book with lots of info tweaking your code.

off to update that other node now...
PS- your benchmark doesnt do anything in either case the line:my $data='for bar baz'; Should be changed to have the char being looked for. Like So: my $data='for-bar-baz';