Your going to have to show us a bit more of your code--like where does %some_el come from?--because on the face of it, 130MB is too big for a hash constructed from 8MB of data.

This creates an 8MB file of keys and values, loads them into a hash, and the total size is just 12MB:

c:\test>perl -E"printf qq[%014d: %014d\n], $_, $_ for 1..262144" >junk
+.dat

c:\test>dir junk.dat
10/04/2010  11:06         8,388,608 junk.dat

c:\test>perl -MDevel::Size=total_size 
-E"local$/; my %h = split ': ', <>; print total_size \%h;" junk.dat
12489744
[download]

Of course, if the 8MB contains more than just a flat hash structure, then the memory requirement will be more, but 10x more is stretching the imagination a bit. So, it probably comes down to what else you are doing in your code. Real code is always more likely to result in a resolution than pseudo-code.