I think a couple of goals have to be served, and no type of training serves all of them.
1. You need to learn how to recognize basic patterns
2. You need to learn how to set up the basic patterns
3. You need to learn how to detect combinations of the basic patterns
4. You need to learn how to set up the combinations
5. You need to improve your speed, and conserve your energy, by honing your intuition about the possibility of combinations
6. You need to develop your depth of (board) vision
To improve pattern recognition and speed, Michael de la Maza advocates a method of repetition for the same large set of tactics problems. Unfortunately, you only learn those problems, and what they represent. After a few repetitions, this has diminishing returns for your board vision as well, since you begin to memorize the lines themselves without thinking about them critically.
So, you need fresh problems as well. This will introduce new combinations, and strengthen your board vision.
A randomizing web site like Chess Tempo can't do the repetitive sets, but it can do the fresh problems well. Unfortunately, I'm finding that I have to discount a fair number of the problems for various reasons.
Also, in a real game, when you sense a tactical combination, you take as much time as you need or as you can afford (whichever is less) before deciding on the first move (unless it's obvious). There is no standard amount of time for a new position, and your available clock time may be completely different than last time you saw the most closely similar position on a board. So, hard and fast timing is not realistic (though it's economical).
Finally, in a real game there are plenty of positions where you are losing or barely equal, and any tactics you suspect just aren't there. Sometimes, it's your job just to detect and prevent an opponent's tactics. Most problem sets completely fail to address these two circumstances (Lev Alburt's little book of 300 Chess Problems is a notable and worthy exception).
I have used CT Art 3.0, and I like it for its ability to serve many needs at once - it offers themes, difficulty levels, practice and test modes, and standard attacking patterns you are really going to see again, like attacking the fianchettoed castle position. Plus, you can see your progress graphically, and get (nominally) rated. And as an earlier reader stated, you get partial credit for partly correct solutions!
Convekta, the makers of CT Art, also offer both beginner's and intermediate tactics offerings, as well as training in the opening (especially traps), strategy and endgame. I couldn't praise their efforts more highly; ICGM Maxim Blokh has done wonders...
By the way, I should point out that Dvoretsky and Nunn's work is not for the faint-hearted, and certainly not for the beginner. If you're already rated 2500 here, you may be overestimating the capabilities of rest of us.
BTW: For the intermediate player I also recommend David Lemoir's new book: How to Become a Deadly Chess Tactician, and Minev's Mastering Tactical Ideas.
Good chess!
