Hey, that's hard to quantify
Basically if you do tests in the book, they don't scale well at all. And I find that with the old paper GMAT tests, your score is usually lower, especially if you're proficient at more difficult problems versus you just being fast at simple problems.
Others have said that GMATPrep 1 & 2, along with PowerPrep 1 & 2, provide pretty much a 1-1 correlation between your test scores and the real scores. So if you get a 700 consistently, you'll probably score 690-710. The formulas they use are somewhat weird. For example I did GMATPrep2 yesterday, and I got 9 Math wrong, and 7 verbal wrong. If you scaled it on an old paper test, I'd get maybe a 660-680, and on a Kaplan, McGrawHill, etc scoring system, I'd get a 600. But because the CAT factors in the difficulty of the questions, I ended up getting 49Q and 45V, which was a 760.
I think it's almost impossible to predict your score other than via the prep software. You have to basically learn all the concepts - data sufficiency, all the math concepts (probabilities, median, average, combinations, etc), grammar rules, how to read fast for paragraph comprehension, and how to do CR properly.
Anyways the problem with the CAT, and predicting your score, is that the first 10 questions determine your level of difficulty. I usually get the first 12-15 right, so I get to the top, then I start getting IMPOSSIBLE questions, which end up taking a lot of time and draining you mentally - but that's what they're trying to do, to see how much you can take. If you get 5/10 of the first ones wrong, then your score will be very bad, even if you get the rest of the questions right. If you get the first 10 right, then get say 18/27 on the last 27, then you'd get a very high score, because the questions would be very difficult.
I would just practise with super hard difficulty questions and focus on speed.
Best of luck