Interestingly, benchmarks do not bear this assumption out. (perl 5.8.0 on a celeron lots of other stuff going on at the same time, but tried multiple times, very similar results).

#!/usr/local/bin/perl
use Benchmark;


for (1.. 100000) {
  $rand = rand * 100;
  push @rvs, $rand;
  $last = $rand;
}
$array[0] = $rand;

timethese ( 100, {grepit => \&grepit,
           hashit => \&hashit});

sub grepit {
  return 1 if (grep /$array[0]/, @rvs);

}

sub hashit {
  @hash{@rvs} = (1) x @rvs;
  return 1 if ($hash{$array[0]});
}
[download]

Benchmark: timing 100 iterations of grepit, hashit...
    grepit: 19 wallclock secs 
       (19.36 usr +  0.03 sys = 19.39 CPU) @  5.16/s (n=100)
    hashit: 28 wallclock secs 
       (27.84 usr +  0.04 sys = 27.88 CPU) @  3.59/s (n=100)
the number is 0.359661571026372
Grepit found the number
Hashit found the number
[download]