This was my first inclination as well -- note that you don't have to sort the file first if you do it this way. A potential downside is that you wind up with the entire data structure in memory.

One option to avoid the memory consumption is to use something like DBD::SQLite as mentioned above. Another option (probably slower and bulkier, but not requiring knowing DBI and SQL) might be to use Tie::MLDBM to store that hash on disk as a DB_File rather than in memory. That might wind up being larger than the original text file, but at least it'll be easy to access it again in perl.

Note: if you use Tie::MLDBM, you'll need to extract the array from the hash to a temporary array, push your data to the temporary array, and then store the temporary array back in the hash. Read the pod for more details. E.g.:

# Code not tested - consider it conceptual #
use strict;
use warnings;
use Tie::MLDBM;

tie my %hash, 'Tie::MLDBM', {
     'Serialise' =>  'Storable',
     'Store'     =>  'DB_File'
 }, 'mybighash.dbm', O_CREAT|O_RDWR, 0640 or die $!;

while (<>) {
  chomp;
  my ($id, $data) = /(\d+)\s+(.*)/;
  my $aref = $hash{$id} || [];
  push @{$aref}, $data;
  $hash{$id} = $aref;
}
[download]

Good luck!