Isn't that statement "not downloading" a bit misleading? Because you do transfer all file contents from S3 to the lambda and back.

Yes, you do not need /tmp storage and download/split/upload happens in parallel but from the data movement perspective it is a download/upload.