I think creation of the BigInt objects might be mucking stuff up. Simplify the test. Loop is faster, and it gets faster as the argument gets bigger.

YuckFoo

Benchmark: timing 16000 iterations of nocurse, recurse...
   nocurse:  0 wallclock secs 
      ( 0.20 usr +  0.00 sys =  0.20 CPU) @ 80000.00/s (n=16000)
   recurse:  1 wallclock secs 
      ( 0.51 usr +  0.00 sys =  0.51 CPU) @ 31372.55/s (n=16000)

#!/usr/bin/perl

   use Benchmark;

   $fact = $ARGV[0] || 10;

   timethese(16000, {'recurse'=>'recurse($fact)',
                     'nocurse'=>'nocurse($fact)'});

sub recurse {
   my ($n) = @_;
   if ($n == 0) { return 1; }
   return ($n * recurse($n-1));
}

sub nocurse {
   my ($n) = @_;
   my $f = 1;
   for my $i (1..$n) { $f *= $i; }
   return $f;
}
[download]