Tar files have a decent amount of overhead. Source: I wrote a streaming untarring library in C for a streaming video product. You would definitely add WAY more than 1 byte of overhead per file, which is what is required for this trick to work.
But it rules out other kinds of tricks like storing info in file names. If all the metadata is counted in the length of the tar file, these tricks don't stand a chance. There's way more than 1 byte of overhead per file in the file system and Mike needed a rule that counts all of them.
