If this were simply finding the number of unique value entries in a flat hash, this would be a fairly straight forward issue - simply use reverse to invert the hash and use keys in either list or scalar context depending on whether you want the list of values or just the count.

However, if what you want to know is how many people are of age 12, as per your example, you have to collect to data by key. Since this means you need to look at each entry, how would you expect to not have to examine each node in a loop? You can use built-ins like map to reduce the total number of characters, but will likely make the code more difficult to read in the future. The clearest way to determine how many people are 12 would be to loop over the hash and collect counts in a hash:

#!/usr/bin/perl
use strict;
use warnings;

my %hoh = (
 kevin => {
   age   => "12",
   favorite_color       => "blue",
   gender       => "boy",
 },
 john => {
   age   => "11",
   favorite_color       => "green",
   gender       => "boy",
 },
 lisa => {
   age   => "11",
   favorite_color       => "pink",
   gender       => "girl",
 },
 sara => {
   age   => "13",
   favorite_color       => "purple",
   gender       => "girl",
 },
 shelly => {
   age   => "12",
   favorite_color       => "purple",
   gender       => "girl",
 },
);
my %counts;
my $key = 'age';
foreach my $person (values %hoh) {
    $counts{$person->{$key}}++;
}
use Data::Dumper;
print Dumper(\%counts);
[download]