If you use the (binary) md5 of the url as the hash key and don't assign anything as the value (ie. use  undef $hash{ md5( $url ) }; to autovivify the key), then storing 10 million urls will require around 1 GB of ram.

If you preallocated enough buckets (keys %hash =  2 **24;), then it runs pretty quickly too.

There is the rare possibility that you will get a false positive by finding two urls that hash to the same md5, but the chances are less than with a bloom filter and if you are using md5 for your Bloom::Filter solution, that would be possible anyway.