Testing on a Solaris box, with the following two scripts:

#script one
$f1="X" x 200000;
$f2=$f1;
$f3=$f1;
...
$f30=$f1;
while (1) {sleep 1}

#script two
for (1..30) {
   $f{$_}="X" x 200000;
} 
while (1) {sleep 1}
[download]

They seem to be identical in memory usage.

Process 1:
Text  VSS: 1.9mb   Data  VSS: 8.1mb   Stack VSS:  16kb
Shmem VSS:   0kb   Other VSS:   0kb   Total VSS:  10mb

Process 2:
Text  VSS: 1.9mb   Data  VSS: 8.1mb   Stack VSS:  16kb
Shmem VSS:   0kb   Other VSS:   0kb   Total VSS:  10mb

I would suspect how you end up populating/using the scalars and/or hashes would have a greater effect.
Comment on Re: Hashes/Scalars and Memory Usage
Download Code