Their measurement isn't wrong. It's demonstrating the exact point in question: that if the logs were just stored in plain text, then grepping them would be an order of magnitude faster (or multiple orders of magnitude in the case of ripgrep) than whatever `journalctl --grep` is doing.