A good idea, tye. Here is the code that proves it. I separated the two parts (create and read file), to prove that the error is in the reading.

First, create the big datafile:

#!/usr/bin/perl

use Tie::Persistent;
my %a = ();
tie %a , 'Tie::Persistent', "bigdatafile" , 'rw';
for my $i (1...5000) {
        $a{"ENTRY$i"}{"has"}{"a lot"}{"of data"};
}

untie %a;
[download]

Now read the big datafile:

#!/usr/bin/perl

use strict;
use Tie::Persistent;

# use this time to look up the process
# and see how much memory it consumes
print "My Process ID is $$\nIn 10 seconds I will tie a big data file:\
+n";
for my $i (1..10) {
    print "$i\n";
    sleep 1;
}
&loadData("bigdatafile");


# now look at memory usage again
sleep 100;

exit 0;


sub loadData {
    my $file = shift;
    my %localhash = ();
    tie %localhash , 'Tie::Persistent', $file , 'rw';
    print "File tied\n";
    untie %localhash;
    print "File untied\n";
    undef %localhash;
}
[download]