use Benchmark 'cmpthese';
use strict;

my %hash = 1 .. 10_000;
my %add = 9_001 .. 11_000;

cmpthese(-5, {
  BASE => sub {
    my %copy = %hash;
  },
  JAPHY => sub {
    my %copy = %hash;
    my @new = grep !exists $copy{$_}, keys %add;
    @copy{@new} = @add{@new};
  },
  ZAXO => sub {
    my %copy = %hash;
    %copy = (%add, %copy);
  },
  NULL => sub {
    my %copy = %hash;
    %copy = (%copy);
  },
});
[download]

        Rate  ZAXO  NULL JAPHY  BASE
ZAXO  15.1/s    --  -22%  -61%  -76%
NULL  19.3/s   28%    --  -50%  -69%
JAPHY 38.9/s  157%  101%    --  -38%
BASE  62.4/s  312%  223%   60%    --
[download]

        Rate  ZAXO  NULL JAPHY  BASE
ZAXO  11.4/s    --  -39%  -58%  -81%
NULL  18.6/s   63%    --  -31%  -69%
JAPHY 26.9/s  136%   45%    --  -56%
BASE  60.5/s  431%  225%  125%    --
[download]

The grep is superfluous and, IMO, only distracts from the code. Here's the sub to add to the above benchmark:

  TANKTALUS => sub {
    my %copy = %hash;
    @copy{keys %add} = values %add;
  },
[download]

And the benchmarks. First, the original 500/500 overlap/new keys:

            Rate      ZAXO      NULL TANKTALUS     JAPHY      BASE
ZAXO      12.3/s        --      -23%      -59%      -59%      -76%
NULL      16.1/s       30%        --      -46%      -47%      -68%
TANKTALUS 29.8/s      142%       85%        --       -2%      -41%
JAPHY     30.4/s      147%       89%        2%        --      -40%
BASE      50.9/s      313%      216%       71%       67%        --
[download]

As you can see, the difference is pretty insignificant compared to the extra "line noise" that needs to be parsed by the human who maintains it. I didn't think it'd change much with a larger sample, but I tried it anyway (using your 5_001..12_000 hash):

            Rate      ZAXO      NULL     JAPHY TANKTALUS      BASE
ZAXO      9.48/s        --      -39%      -55%      -55%      -81%
NULL      15.7/s       65%        --      -26%      -26%      -69%
JAPHY     21.1/s      122%       35%        --       -0%      -59%
TANKTALUS 21.1/s      123%       35%        0%        --      -59%
BASE      51.1/s      439%      226%      142%      142%        --
[download]

Somewhere, deep down in the lost precision, the non-grep version comes out ahead. Which we'll attribute to statistical insigificance, just as I did for the 2% in the other direction. ;-)

The other point is that your answer does not do the same thing as the rest of the examples here. In all the rest of the cases, the new hash always overrides the values in the old hash. That is, if the old hash had a => "b", while the new hash had a => "c", the behaviour of yours vs the rest is different. Which one is correct is dependant on the demands of the situation, although I usually lean towards new hashes winning over old hashes the way I write the code. In cases where I want "first come, first served" policies, then I will definitely keep your grep in mind - it's probably a lot more elegant than the way I have been doing it (if I have - I'd have to look through old code to be sure). Thanks ;-)

You missed the point, Tanktalus. We're only adding NEW key-values pairs to the hash, thus the grep() is required. The node by Zaxo was showing how to have the original values override the new values if the keys matched.
_____________________________________________________
Jeff japhy Pinyan, P.L., P.M., P.O.D, X.S.: Perl, regex, and perl hacker
How can we ever be the sold short or the cheated, we who for every service have long ago been overpaid? ~~ Meister Eckhart