Here's a not-thoroughly-tested | moderately-well-tested recursive approach for any reference to a hash/array-of-hash/array structure of any depth. Whether this is "nice" is perhaps debatable, but at least it avoids a for-loop.

c:\@Work\Perl\monks>perl -wMstrict -MData::Dump -le
"my $ComponentData = {
   'key_A' => {
     '6.0' => [
       { 'FILENAME' => 'filename',  'VERSION'  => 'version',  ooo => '
+xxx', },
       ],
     },
   };
 dd $ComponentData;
 ;;
 print highest_multi_element($ComponentData)->{FILENAME};
 print highest_multi_element($ComponentData)->{ooo};
 my $coderef = sub { return { ooo => 'yyy' }; };
 print highest_multi_element($coderef)->{ooo};
 ;;
 sub highest_multi_element {
   my $dataref = shift;
   ;;
   if (ref $dataref eq 'HASH') {
     my ($single, $val, $multi) = %$dataref;
     die 'empty hash' unless defined $single;
     return defined $multi ? $dataref : highest_multi_element($val);
     }
   elsif (ref $dataref eq 'ARRAY') {
     my $items = @$dataref;
     die 'empty array' unless $items;
     return $items > 1 ? $dataref : highest_multi_element($dataref->[0
+]);
     }
   else {
     die qq{not hash/array ref: $dataref};
     }
   }
"
{
  key_A => {
    "6.0" => [
      { FILENAME => "filename", ooo => "xxx", VERSION => "version" },
    ],
  },
}

filename
xxx
not hash/array ref: CODE(0x66a6bc) at -e line 1.
[download]

Update: Changed code example as follows:

• Changed function name from  lowest_multi_element() to  highest_multi_element() because we're actually looking for the highest level multi-element hash/array ref.;
• Placed the test for a hash reference first because these refs may be (?) more common;
• Used LanX's hash-to-list-of-scalar-lexicals-flattening hack because it may be slightly faster than extracting all values to an array — and also because I think it's just cooler.

I've also had a chance to do more testing and I'm more confident about the robustness of the function.