Two points:

1. I wouldn't worry about the efficiency until you get it working right. Doesn't matter how slow it is if it's incorrect.
2. Why does it have to be a regex?

#!/usr/bin/perl -w

use strict;

my $test = "alpha beta  beta gamma gamma gamma delta";

my %seen;
my @out;
for ( split( " ", $test ) ) {
        push( @out, $_ ) unless $seen{$_}++;
}

print join( " ", @out ), "\n";
[download]

This solution keeps the order, dedupes throughout the entire string, and doesn't use any regexes, and I'd bet it's a good deal quicker than any regex solution. Of course, you'll want to measure if you want to be sure.

xoxo,
Andy
--
<megaphone> Throw down the gun and tiara and come out of the float! </megaphone>