If you flatten your data structure, it would be a lot easier to access the sort key wanted. Of course this is not optimal as you need to duplicate all your data in a temporary array.

#! /usr/bin/perl

use strict;
use warnings;

my %h;

# if your data is really numerical then the apostrophes are superfluou
+s
$h{10}{20} = 5;
$h{10}{25} = 4;
$h{05}{50} = 3;
$h{20}{30} = 2;

my @temp;

for my $key1 ( keys %h ) {
  for my $key2 ( keys %{ $h{$key1} } ) {
    # create a 3-column table from your hash
    push @temp, [ $key1, $key2, $h{$key1}{$key2} ];
  }
}

# sort it by the second column numerically
for ( sort { $a->[1] <=> $b->[1] } @temp ) {
  print 'key1=', $_->[0], ', key2=', $_->[1], ', value=', $_->[2], "\n
+";
}
[download]

I hope this helps.

update: a little clarification: By using the solution above you need to store every one of your key1, key2 and value at least once more, but in the terms of memory consumption this can be more than a duplicate.