You'd think so, but then every other calculation in that paragraph is based on bytes: 1MB for 8192 identifiers (as opposed to 128K as strings, or 64K as ints)

You'd think so, but then every other calculation in that paragraph is based on bytes: 1MB for 8192 identifiers (as opposed to 128K as strings, or 64K as ints) 384GB for 3 billion identifiers (rather than 38GB strings, or 19GB ints) 119GiB if 1 in 3 is unique (rather than 12GiB strings, or 6GiB ints) 6GiB is not a lot of RAM anymore.