Thanks for your comments.
I had the feeling that not a single question on this test was solvable by dimensional analysis
If you look at the most recently released test from 2008, you'll see that with one exception, no question on this exam is entirely solvable by dimensional analysis either (the graph reading question #41 being the exception). Indeed, the trend in recent GRE's seems to allow you to eliminate a couple answer choices by dimensional analysis (GRE '08 questions 7, 10, 12, etc), or to forbid dimensional analysis by giving answer choices the same units (22, 26, 40, etc). Our tests try to reproduce this pattern. However, there are
quite a few questions on our Sample Exam 1 which can be solved by just a combination of dimensional analysis and limiting cases: 39 and 84 in particular.
and that there was a more-than-usual amount of "topical" questions like astrophysics and so on
Our exams reproduce the content distribution given by ETS to within +/- 1 question. You may be right that we had more astro questions than condensed matter in the special topics section, but for special topics, really anything is fair game and you'll get hit with a good deal of random trivia on the real exam.
Also, ETS tends to put questions in "packets" of same subject, which makes it easier compared to a random order like yours.
The "packets" you talk about are typically 2-4 questions in length drawn from the same subject area. Looking back at our exams, you're right that we tend to stick to blocks of 2, rather than 3 or 4. But no one says you have to do the questions in order! You're free to go through the exam only doing classical mechanics, then E+M, or however you choose.