That won't fly.

For one, sort reads its entire input before outputting a single character, so the sort process will grow not only comparably to the hash inside the perl process, it will infact grow larger than the entire input file. It does considerably more work too - the single-pass approach doesn't need to sort the data since it uses a hash to keep words unique.

Secondly, the hash you're creating in the second pass is exactly as large as the hash would be at the end of the single-pass script - they both contain all unique words in the file. But you create that hash before you start processing, so your second pass will start out with as much memory consumed as the single-pass scripts reach only by the end of their processing.

Makeshifts last the longest.