I just got done playing with his code. I took out the multiple calls to srand (now a single call: srand(iseed);), the do nothing for loop and various other oddities ((2 / 1) ?). It now creates very regular terrains. The generated terrains show a very strong rotational symmetry. Each 1/4 of the image is nearly an exact copy of each other, rotated 90°.
I think the entire algorithm needs to be rewritten.