perl -le '@l=sort{rand$_<=>rand$_}(1..10);print for@l[0..2]'
[download]

In older Perls, it's not garanteed to give any meaningful results - in fact, perl might as well segfault. In modern Perls, about all that's garanteed is that perl won't segfault, and perl won't internally get corrupted. I'm not at all convinced that the above code produces fair results. Here's a test with a smaller array that strongly suggests your algorithm isn't fair:

#!/usr/bin/perl

use strict;
use warnings;
no warnings qw /syntax/;

my @array = (1 .. 4);
my %count;
my $mult  = 10_000;

for (1 .. 24 * $mult) {
    my @t = sort {rand 1 <=> rand 1} @array;
    $count {"@t"} ++;
}

foreach my $key (sort keys %count) {
    printf "%5d: %s\n" => $count {$key}, $key
}

__END__
15063: 1 2 3 4
15069: 1 2 4 3
 7419: 1 3 2 4
 7615: 1 3 4 2
 7602: 1 4 2 3
 7606: 1 4 3 2
14978: 2 1 3 4
14923: 2 1 4 3
 7497: 2 3 1 4
 7460: 2 3 4 1
 7558: 2 4 1 3
 7401: 2 4 3 1
 7485: 3 1 2 4
 7484: 3 1 4 2
 7585: 3 2 1 4
 7577: 3 2 4 1
14876: 3 4 1 2
14968: 3 4 2 1
 7467: 4 1 2 3
 7544: 4 1 3 2
 7461: 4 2 1 3
 7477: 4 2 3 1
14853: 4 3 1 2
15032: 4 3 2 1
[download]

Abigail