For this particular test I ran, hash assignment and slices seemed to be much faster than creating the union one item at a time:

use strict;
use warnings;
use Benchmark 'cmpthese';

my (%set_a, %set_b);
@set_a{1..100} = undef;
@set_b{50..150} = undef;

sub slices {
  my %set;
  @set{keys %set_a, keys %set_b} = undef;
}

sub items {
  my %set;
  $set{$_} = undef for keys %set_a, keys %set_b;
}

sub hasheq {
  my %set;
  %set = (%set_a, %set_b);
}

cmpthese(10000, {
  slices => \&slices,
  items  => \&items,
  hasheq => \&hasheq,
});
[download]

gives these results on my machine:

Benchmark: timing 10000 iterations of hasheq, items, slices...
    hasheq:  3 wallclock secs ( 2.97 usr +  0.00 sys =  2.97 CPU) @ 33
+67.00/s (n=10000)
     items:  4 wallclock secs ( 4.18 usr +  0.00 sys =  4.18 CPU) @ 23
+92.34/s (n=10000)
    slices:  3 wallclock secs ( 2.96 usr +  0.00 sys =  2.96 CPU) @ 33
+78.38/s (n=10000)
         Rate  items hasheq slices
items  2392/s     --   -29%   -29%
hasheq 3367/s    41%     --    -0%
slices 3378/s    41%     0%     --
[download]

-- Mike

--
just,my${.02}