Extractor uses a hash table to keep track of previously encountered strings to only store them once. Unfortunately, that table uses a lot of RAM on large extracts.