To avoid having one foreach per level in the hash, I would write a recursive hash traversal subroutine as in this (tested) example, which uses a single foreach in a subroutine that calls itself when it finds a hash reference instead of a value:

Update 1: the point of all this is not how many times foreach is actually executed but to avoid having to code differently for different nesting levels of the hash.

Update 2: modified to show how to make the actual processing of the hash flexible. The assignment into a flat array is just to demonstrate that all the values and none of the keys of the hash were covered. A slight change can make it do something else at each node.

Update 3: where the code examines ref($val), it could also be adapted to process arrays within hashes, by creating an extra branch for where ref($val) eq 'ARRAY'.

#!/usr/bin/perl

use strict;
use Data::Dumper;

my %h = ();

# stick in some data at varying depths
$h{ 1}{ 2}{ 3} = 123;
$h{1}{3} = 13;
$h{2} = 2;

# this array will collect the values
my @flat = ();

hashTraverse( \%h, \@flat );

print Dumper( \@flat );

sub hashTraverse {
    my ( $href, $aref ) = ( shift(), shift() );
    foreach my $key ( sort keys %$href ) { # this simple example sorts
+ the tree left-right
        my $val = $href -> { $key };
        if ( ref( $val ) eq 'HASH' ) {
            hashTraverse( $href -> { $key }, $aref )
        }
        else {
            push @$aref, $val;
            # or instead whatever else should be done to or with the h
+ash node
        }
    }
}
[download]

Output:

$VAR1 = [
          123,
          13,
          2
        ];
[download]