The tool of choice back then was SoftICE and it would have been trivial to trap even bios level LPT access.

More to the point the guy didn't even open the dongle, which in all likelihood was a simple set of logic gates in this case.

I'm assuming he did have a careful look at the caller to the function because his reasoning that the communicating function took no inputs is faulty.

What if the hardware has a LFSR in it and returns changing responses, and the caller can check them with a verify function else where that maintains a state and knows to expect the next in the sequence?

We have far better tools today for sure, but we understood the system better back then. The right tool is sometimes better than a sharper tool as they say.

Edit : I should add I'm aware it's very likely the actual reversing involved more steps and checks than Dimitrys blog which is narrowed to what worked.