Wild and ugly plan follows for comments...

If the code below gives me the memory footprint of the perl interpreter,

my $mem1 = `perl -e 'print \`ps h -o sz \$\$\`'`;
[download]

and the code below gives me the footprint of this mini script

my $mem2 = `perl -e '\$a = {A=>1, B=>2, C=>3}; print \`ps h -o sz \$\$
+\`'`;
[download]

Then shouldn't $mem2 - $mem1 = rought overhead of $a ? full test:

my $mem1 = `perl -e 'print \`ps h -o sz \$\$\`'`;
my $mem2 = `perl -e '\$a = {A=>1, B=>2, C=>3}; print \`ps h -o sz \$\$
+\`'`;
print "$mem2 - $mem1 = ". ($mem2 - $mem1) . "\n";
[download]

Feel free to say "why don't you use module xy instead ?"... I would much rather do just that :)

Tiago