use strict;
use warnings;

my %data = ( 1 => { a => 10,
                    b => 5,
                    c => 20 },
             2 => { a => 90,
                    b => 15,
                    c => 40 } );

my @top_3 = (
             sort { $a->[-1] <=> $b->[-1] }
             map { my $k = $_;
                   my $h  = $data{ $_ };
                   map [ $k, $_, $h->{$_} ], keys %$h
                 } keys %data
            )[ 0..2 ];

for ( @top_3 ) {
  printf "\$data{%d}{%s} = %d\n",
    $_->[0], $_->[1],
    $data{$_->[0]}{$_->[1]};
}
__END__
$data{1}{b} = 5
$data{1}{a} = 10
$data{2}{b} = 15
[download]

As ysth says your question is a bit vague. Anyway, here's an approach that you can start to work with.

use Hash::Flatten qw/ flatten unflatten /;

sub pair { ... } # 1,2,3,4 => [1,2],[3,4]

my %data;
$data{1} = {'a' => 10, 'b' => 5, 'c' => 20 };
$data{2} = {'a' => 90, 'b' => 15, 'c' => 40 };

my @sorted = sort { $a->[1] <=> $b->[1] } pair(%{flatten(\%data)});
print $_->[0], "\n" for @sorted[0..2];

__END__
1.b
1.a
2.b
[download]

ihb

Update: I see other's interpreted "top" as meaning lowest; I was somehow stuck on having "top" mean highest, and had to somewhat unnaturally interpret "take" as "exclude" :)

 $data{1} = {'a' => 10, 'b' => 5, 'c' => 20 };
 $data{2} = {'a' => 90, 'b' => 15, 'c' => 40 };

  @Top =
     sort { $a <=> $b }
        map {
# cut
                        values %{$data{$_}}
          } keys %data;
  @Top[0..2];

# or simplify to
  (
      sort { $a <=> $b }
        map {
# cut
                        values %{$data{$_}}
          } keys %data;
     )[0..2];
[download]

Why are you doing two sorts or looking at keys? All you work with is the values of the values:

my  @Top =
     sort { $a <=> $b }
        map { values %{$_} } values %data;
[download]

---

Caution: Contents may have been coded under pressure.

Thanks. But I need the key values too. Anyways , got the solution.

Thanks again,
Swaroop