But still doing a print call gives overhead,
because the print buffer will be flushed much
more often.

use Benchmark::Timer;

my $t = Benchmark::Timer->new();
my $size = 1000000;

$t->start( 'string' );
open FW, "> temp.out";
print FW " " x $size;
close FW;
$t->stop( 'string' );

$t->start( 'loop' );
open FW, "> temp.out";
for ( 0..$size ) { print FW " " }
close FW;
$t->stop( 'loop' );

$t->report;
__END__
Reports:
1 trial of string (17.594ms total)
1 trial of loop (733.321ms total)
[download]

This clearly shows the loop decreases performance.

Update: I figured that perl could create the string
at compile time, so I put the size in a variable. But
this yields thesame results. Perl still optimizes
the string version in this exapmle. But that was
also the point I made in the first reply.