Here's a short compare in benchmark terms of the two solutions. I added a third alternative using index - which can only be used for fixed strings whereas the other solutions can also use regexes. The following code was used:

use strict;
use warnings;
use Benchmark;

our $cnt = 0;
our $foo = q/abcdef/ x 10_000;

Benchmark::cmpthese( 1000, {
    'while' => sub {$cnt++ while $foo =~ /abc/g;},
    'subst' => sub {$cnt = $foo =~ s/abc/abc/g;},
    'index' => sub {my $POS = -1;
        $cnt++ while (($POS = index($foo,'abc',$POS+1)) > -1);},
});

# prints
Benchmark: timing 1000 iterations of index, subst, while...
     index: 27 wallclock secs (25.09 usr +  0.02 sys = 25.11 CPU) @ 39
+.83/s (n=1000)
     subst: 15 wallclock secs (14.93 usr +  0.02 sys = 14.95 CPU) @ 66
+.89/s (n=1000)
     while: 18 wallclock secs (17.08 usr +  0.01 sys = 17.09 CPU) @ 58
+.53/s (n=1000)
        Rate index while subst
index 39.8/s    --  -32%  -40%
while 58.5/s   47%    --  -12%
subst 66.9/s   68%   14%    --
[download]

The results vary with the length of the pattern searched for. The longer the pattern, the slower the substitution solution (as it has to substitute a longer string). The index version is always slowest. For the case where there are zero occurences of the pattern, all three alternatives take approx. the same time.

-- Hofmator