Comparative benchmarks like this help settle numerous disagreements, and I appreciate and applaud the work that goes into them. Still, it does help if the code fragments being benchmarked do the same thing. The fragments tested do not. Consider
$sum=0; for (my $i=0;$i++<=$n;) {$sum+=$i}
Side-effecting $i prematurely breaks the calculation (an off-by-$n error).