Hi! I recently started to play with both the Lego official software for GyroBoy and the Dean Hysted one you were quoting.
I run them on a slightly modified version of the Lego GyroBoy, since I had all the sensors, but not all the construction parts and both do work smoothly.
From what I understood, Dean did a great job understanding Lego software and "simply" wrote a new version with exactly the same algorithm, but in much better, more readable, more structured program which is easily undestood.
He used the same parameters taken from Lego, even though, even he had not clear where those numbers came from.
From the little I've seen, the underlying theory should be some proportional-integral-derivative-controller (PID-controller) with 4 parameters based on gains for gyro angle, gyro rate, motor angle and motor speed.
Hope that will help a little
