In this particular case, the change is easy. Instead of using $$href{$word}{written}, you can use $href->{word} = 'written', because there are no interesting values in the second level hash.

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

my %hash = ( 1 => { one   => q() },
             2 => { two   => q(), zwei => q(), dos => q() },
             3 => { three => q(), tres => q() },
             4 => { four  => q() },
             5 => { five  => q() },
             6 => { six   => q(), seis => q() },
             7 => { seven => q() },
           );


my $valCount;
for my $item (keys %hash) {
    my $count = keys %{ $hash{$item} };
    print "count is $count\n";
    $valCount += $count;
}

my %subs;
my $recCount;
while ($valCount > 0) {
    for my $id (sort keys %hash) {
        if (not exists $subs{$id}) {
            my $href = $hash{$id};
            for my $word (sort keys %$href) {
                next if 'written' eq $href->{$word};

                $valCount--;
                print "$id $word\n";
                $subs{$id} = q();
                $href->{$word} = 'written';
                last if keys %{ $hash{$id} } > 1;
            }
            $recCount++;
            if ($recCount == 4) {
                $recCount = 0;
                %subs = ();
            }
        }
    }
}
[download]

Generally, you'd need to introduce one more layer to store the original value and the new hash:

my %hash = ( 1 => { one => { value => 'v' }, ...
[download]