@common = qw|a an and at the to|;

timethese(1000, {
  Odud => q{
    my (@only, %r, %seen);

    foreach $r (@words){
      $r{lc $r} = 1;
    }
    my @uniqwords = sort keys %r;

    foreach my $item (@uniqwords) {
      push(@only,$item) unless exists $seen{$item};
    }
    my @newwords = @only;
  },
  PMGurus => q{
    my %seen;
    @seen{@common} = (1) x @common;
    my @newwords = sort grep !$seen{+lc}++, @words;
  },
  btrott_OneLiner => q{
    my %seen;
    @seen{@common} = (1) x @common;
    my @newwords = sort grep !$seen{$_=lc}++, @words;
  },
  btrott_FirstPass => q{
    my (%seen, %r);
    @seen{@common} = (1) x @common;
    for my $r (@words) {
      next if exists $seen{ lc $r };
      $r{lc $r} = 1;
    }
    my @newwords = sort keys %r;
  },
  visnu => q{
    my %common;
    @common{@common} = (1) x @common;
    my @newwords = sort keys %{+{ map { !$common{$_ = lc $_} ? ($_, 1) : () }
                                       @words }};
  },
});

Benchmark: timing 1000 iterations of Odud, PMGurus, btrott_FirstPass, btrott_OneLiner, visnu...
           Odud:  9 wallclock secs ( 9.61 usr +  0.00 sys =  9.61 CPU)
        PMGurus:  6 wallclock secs ( 5.40 usr +  0.00 sys =  5.40 CPU)
btrott_FirstPass: 7 wallclock secs ( 7.07 usr +  0.00 sys =  7.07 CPU)
btrott_OneLiner:  6 wallclock secs ( 5.73 usr +  0.00 sys =  5.73 CPU)
          visnu: 10 wallclock secs (10.29 usr +  0.00 sys = 10.29 CPU)