I suppose they are streaming the tarball directly to S3 using multipart uploads, maybe parallelizing this task at the same time.
You wouldn't be able to do it using regular tar.
You wouldn't be able to do it using regular tar.
For very large uploads you must also specify a chunksize using "multipart_chunksize" such that the upload will require 10,000 chunks or fewer (the CLI can't do this for you because it doesn't know the size of your stream up front).