Benchmark overhead of copying hash reference to a new variable, four times...

benchmark_hash_by_ref.pl

#!perl

use warnings;
use strict;

use Benchmark qw( cmpthese );

my $h0 = { 
    'h1' => {
        'h2' => {
            'h3' => {
                'h4' => {
                    'key' => 0,
                },
            },
        },
    },  
};

cmpthese(100_000_000, {
  'use_direct' => sub { sub_direct() },
  'use_ref' => sub { sub_ref() },
});

sub sub_direct {
    $h0->{'h1'}{'h2'}{'h3'}{'h4'}{'key'} = 1;
}

sub sub_ref {
    my $h1 = $h0->{'h1'};
    my $h2 = $h1->{'h2'};
    my $h3 = $h2->{'h3'};
    my $h4 = $h3->{'h4'};
    $h4->{'key'} = 1;
}

                            Rate    use_ref use_direct
            use_ref    2656042/s         --       -39%
            use_direct 4378284/s        65%         --

Makes sense... Copy ref to the hash into new variable along with the memory allocation for that new variable.
Should make a variant where I have multiple sub_ref like functions that copy less and less references to check if that is linear.