Well, I am a member of a Russian computer chess forum. Many of its members do various tests on their monster machines - I7 x64 Win and alike.. But all these tests are on short time controls, i.e. irrelevant to the real power of the known chess engines. So to speak, not one of these members can state which engine is best for a long time analysis. They say that only a test in not less than a 1000 games can be real. What is more, such test has to go on a longer time control than just for 2 hours for 40 moves or so. And I think they're right.

The fact that Houdini 2.0c (the last version) is best on short time controls doesn't mean that this engine is best for analysis at all. And all the rating lists are based on games played on short time controls.
I think one has to analyse with more than one engine - for example: Houdini 2.0 and Rybka 4.1 because they seem to have different programming codes. What I mean is...if I analyse with Houdini 2.0 and Critter 1.4a (which seems to be a very good engine, too) both give me almost the same moves as best. While Ivanhoe, Komodo and Stockfish are different. Some say Rybka is a clone of Fruit 2.1 though improved.. Anyway, one has to find two engines that give different moves as best in a position in say 25 plys depth and experiment with the results. After all, the human knowledge will prevail in giving the best evaluation.

As for the openings all of these engines are of no help except in some specific cases when tactics prevail. And no engine can be useful for a good evaluation of an endgame if there are no endgame tablebases installed on your computer. One has to know that the endgame tablebases have to be activated in the middlegame analysis where the chess engines often count long lines till deep endgames. Some correspondent chess players forget it and often struggle for that.

What is known for now:
1. Houdini 2.0c is best on short time controls. It counts fast and gathers depth in plys very fast. It prunes lines recognized as bad better than others, i.e. its speed.
2. Houdini 1.5a (free engine) - almost as good as Houdini 2.0c. I think it's better for analyses because it doesn't prune so easy lines recognized as bad.
3. Critter 1.4a (free engine) - some say it's better in closed positions and good at endgames without tablebases. It counts very fast too, but not as fast as Houdini 2.0.
4. Rybka 4.1 is finishing the Big Four as the most balanced one in all kind of positions. Some say when it states depth 18 it means depth 21 on other chess engines (i.e. +3 plys real depth).
5. Komodo 4 (goin' commercial) is the newest version and some say it deserves attention because of Larry Kaufman's original dynamic evaluation of pieces' activity.
6. Strelka 5.1 (free engine) - Yuri Osipov's no multi-variations' engine. Very fast and somehow original but with wrong digit evaluations (for example +1.47 in a += position) which can be trusted only for a suggestion of the best move. Some say it's rewritten Rybka from inside out.

7. Stockfish 2.2 or Ivanhoe 9.. - both good for a second analysing engine.
8. The new Robbolito 0.10 SMP - the last version of a legendary free engine - a programming (Ippolit) code which was a source for many of above-mentioned and alike. Now good as a second analysing engine, too. It doesn't use endgame tablebases though and the author says it is improved in endgames.

Anyway, my modest recommendation is to use Houdini 1.5a as a first analyzing engine (as I said it doesn't prune lines as easy as Houdini 2.0c) and Rybka 4.1 as a second analysing engine. And in some positions you can try a third (I can say - original) engine like Strelka 5.1 or Komodo 4 or Robbolito 0.10 just to look for a new idea or something..