as for the first point, yup, I understood that from the article; I probably should have used "similar to Steam" rather than "like Steam". I mostly meant, you'd get a long stream of characters and you'd have to manually try to dejumble them.

Whereas, I believe if you go by your second point, you can see "Ok, the user put username <x> in the username textfield, password <y> in the password field, address <a> in the address textfield, port <p> in the port textfield" and so on, which would make for a more structured data dump. Maybe not possible or feasible for every single application, but if you could get the highest usage targets, like the most common FTP clients, or Steam as they have apparently done, and the browser password storage stuff (or fields for say, most common banking sites, PayPal, etc.) then you could save yourself a lot of time.

Most of them usually show specifics, the keylog file is often less of a jumble and more of like:

{TAB}{TAB}{CURRENT WINDOW: Steam - Log in}{TAB}username{LEFTCLICK}password{LEFTCLICK}{CURRENT WINDOW CHANGED: Steam: Home}

That's not terrible. It would be fairly easy, almost trivial, to write a Perl/Ruby/... script to capture a large chunk of useful credentials.