What's Better: Doing Many Questions on 1 Topic or a Mix??

a couple of suggestions.

* WHEN YOU STILL HAVE LOTS OF TIME UNTIL THE TEST DATE,
you should isolate the areas that are your weakest, and practice lots of problems in those areas at once. the goal is not to accumulate tons of experience indiscriminately, though; rather, the goal is to MAKE CONNECTIONS between those problems.
in other words, as you work through multiple problems dealing with the same topic, you should be paying attention to what those probems have in common more than to the niceties of the solutions themselves. that way, you'll be better equipped to recognize similar problems, and know which approaches to use, when they appear on the real test.

* WHEN THE TEST DATE DRAWS CLOSER,
you should be starting to attack random chunks of problems, so that you have a better simulacrum of the real testing experience (and so that you can't derive an artificial benefit on later problems just because you've been working lots of problems about the same topic).
when you attack these random chunks of problems, the goal is to have an IMMEDIATE approach to the problem. you don't have to be able to think your way through the problem completely - very few students can do that, even at the top score levels - but you have to know how to START the problem.
if you find yourself staring helplessly at a problem, simply not knowing any approach you can take to it, then that's the point at which you want to go back and look at similar problems (so that you can make the necessary connections). otherwise, at this point you should be doing chunks of unrelated problems, just like the real thing.

i'm not sure what's available from other companies, but note that our manhattangmat strategy guides offer lists of OG problems, organized by content area, in the back of each guide.
the o.g. problems themselves are arranged in rough order of difficulty (with more exceptions in verbal than math), but are not arranged in any manner whatsoever with respect to strategy, topic area, or content.