Unfortunately, your benchmark is wrong. As the array @unsorted is lexical, it isn't visible within the benchmark code, so the subs are uniquifying an empty array. All your benchmark is measuring is how quickly the sub return nothing. Correcting that, I get these results.

#!perl -w

use strict;
use Benchmark ':all';

our @unsorted = map {
    join('', map { ('a'..'z','A'..'Z',0..9)[rand 62] } 1..50)
} 1..5000;

sub uniq_dragonchild { my %x; @x{@_} = @_; values %x }
sub uniq_BrowserUk { my %x; @x{@_} = (); keys %x }
sub uniq_Zaxo { keys %{ { map { $_ => undef } @_ } } }
sub uniq_Roy { my %x; grep {!$x{$_}++} @_ }

our( @a, @b, @c, @d );

cmpthese( -1, {
        uniq_dragonchild  => '@a = uniq_dragonchild( @unsorted )',
        uniq_BrowserUk    => '@b = uniq_BrowserUk( @unsorted )',
        uniq_Zaxo         => '@c = uniq_Zaxo(@unsorted)',
        uniq_Roy          => '@d = uniq_Roy( @unsorted )',
});

#my $a='a'; print $/, $a++, ' : ', "@{[ sort @$_ ]}" for ( \@a, \@b, \
+@c, \@d );

__END__
P:\test>363760
                   Rate uniq_dragonchild   uniq_Zaxo     uniq_Roy uniq
+_BrowserUk
uniq_dragonchild 18.1/s               --        -21%         -31%     
+      -41%
uniq_Zaxo        22.8/s              26%          --         -13%     
+      -25%
uniq_Roy         26.3/s              45%         15%           --     
+      -14%
uniq_BrowserUk   30.6/s              69%         34%          16%     
+        --
[download]