While this has been shown to be less than optimal in benchmarks on other nodes, it is the simplest to implement. However, it should be noted that (as far as I am aware) the behavior regarding the merger of duplicate hash keys is undefined. On my computer it happens to be right-associative:

my %a = ('c'=>1,'d'=>2);
my %b = ('d'=>1,'e'=>2);

my %ab = (%a, %b);  # 'd' => 1
my %ba = (%b, %a);  # 'd' => 2
[download]

If you need to know (or specify) the behavior regarding duplicate keys as mentioned in the OP, you'll probably need a map in conjunction with one of the LIST context examples below.

UPDATE: For example, to create a LIST of repeated key values:

my %a = ('c'=>1,'d'=>2);
my %b = ('d'=>1,'e'=>2);

my @b{keys %a} = map { exists $b{$_} ? [$b{$_}, $a{$_}] : $a{$_} } key
+s %a;  # 'd' => [1, 2]
[download]

Another way:

@my_hash{keys %my_second_hash} = values %my_second_hash;
[download]

The keys from %my_second_hash will clobber any duplicates that are in %my_hash.

My very rudimentary non-conclusive damned lies recommend that the version you showed here is a fair bit faster than either creating the third hash or the %my_hash = ( %my_hash, %my_second_hash ); variant.

-- Douglas Hunter

#!/usr/bin/perl -w
use strict;
use Data::Dumper;

my %hash1 = ('key1' => 'val1', 'key2' => 'val9');
my %hash2 = ('key3' => 'val10','key1' => 'val2');

while (my ($key, $val) = each %hash1)
{
    $hash1{$key} = [$val];
}

while (my ($key, $val) = each %hash2)
{
    push @{$hash1{$key}}, $hash2{$key};
}

print Dumper \%hash1;
[download]

$VAR1 = {
          'key2' => [
                      'val9'
                    ],
          'key1' => [
                      'val1',
                      'val2'
                    ],
          'key3' => [
                      'val10'
                    ]
        };
[download]