But... what if the DB file is really big? (Sometimes people tie a hash to a DB file because of the amount of data, not just for persistence.) When I tried the following test:

perl -MDB_File -MData::Dumper -e 'tie %h, "DB_File", "junk.db"; 
for ($i=0; $i<1_000_000; $i++) { $h{"key_$i"}="value_$i" };
warn "the hash is loaded\n"; sleep 15; warn "starting dump\n"; 
print Dumper(\%h); warn "the hash has been dumped\n"; sleep 15' > /dev
+/null
[download]

Memory usage stayed at about 27 MB (macosx/perl 5.8.6) while the hash was being built and throughout the first sleep, then climbed over 450 MB during the Dump phase. Data::Dumper was making its own internal copies of the keys and values.

(The "junk.db" file itself was 47 MB, and a plain-text print out of keys and values as I suggested above would probably be about half that.)