I have a relatively unique view on DKP systems, the different loot distribution systems using some sort of points to determine who gets what loot. The average WoW player is not very interested in math and statistics, but is very interested in getting epics. I happen to be good at math, and for me epics are not the most important thing in going raiding. That allows me to take a step back and look at DKP systems from a more statistical point of view, what is a particular system likely to achieve, as opposed to "what's in it for me". I already talked about the principal problems in an earlier post.
The DKP system that most people who compared different systems consider to be the most fair is the zero-sum DKP system. It has definitive advantages in evening out the distribution of epics. Statistically speaking, that means over a large number of raids, two players participating in the same raids will end up with the same number of epics, while two players where one plays twice as much as the other will end up with the player participating more getting exactly twice as many epics as the other.
Unfortunately the zero-sum DKP system also has disadvantages. Imagine a guild not being able to beat Onyxia / Ragnaros / whatever boss yet and scheduling raids to repeatedly try to beat that boss, with a low chance of success. You sign up several evenings, each evening you wipe against that boss several times, you never succeed, but you advance the knowledge of the guild in the correct strategy and coordination. In a zero-sum DKP system you get absolutely no points for that effort. If you work out the strategy for Onyxia in 5 evenings of wipes, only beating her on the last attempt, a player participating in all 5 raids gets exactly as many points as a player only present for the last, successful attempt. That seems unfair. A zero-sum DKP system earns you more points for stupid farming runs than for trying to achieve raid progress for the guild among sweat and tears. Also a zero-sum system is unable to integrate reward points for desirable behavior (like being on time) or negative points for undesirable behavior (like not showing up when you promised to), unless you can come up with a complicated formula that balances rewards and punishment to still result in a zero sum.
So many guilds run systems which are not zero-sum. Obviously a system where the sum of points is negative, that is after a number of raids the average DKP score is negative, wouldn't work very well. A negative-sum system would favor people who go on raids less often over people who go more often, which is counterintuitive. As a result the majority of DKP systems is positive sum.
The advantage of a positive sum system is that besides giving people points for every boss killed, you can award them points for valiantly getting repeatedly slaughtered by a new boss, for being punctual, for bringing a field repair bot that helps the whole raid, and for whatever other things you want to reward. And you can even hand out penalties for bad behavior, although you obviously shouldn't overdo that like the famous Onyxia raid leader Dives handing out -50 DKP left and right.
Unfortunately such a system works well for a couple of weeks, but over the long run the disadvantages of a positive sum system become apparent. The further away the sum of points earned minus points spent for epics deviates from zero, the faster the system becomes a source of problems. Imagine an epic item costs 100 points, but on the average 40-man raid every member receives 100 points, while they find 20 epics. That means that on that raid everybody gets the points to "buy" one epic, but only half the players can spend their points, while the other half can only accumulate the points. As long as the same 40 people participate in every raid, that is no problem at all. But we all no that this practically never happens, due to Real Life ® constraints. Two players of the same class, with one player participating twice as often as the other, with the point distribution explained above ends up with the player playing more often getting *ALL* the items, and the player playing less often getting nothing. That is because in two raids the first player accumulates rights for two items, and the second player only for one item. As in two raids only 1 items is found for them (statistically), the first player always has more rights than the second and always gets first choice.
Of course if an item drops that the first player already has, he passes and the second player gets it. Now the result of this in practice is that you get a ranking of players of each class, with the most frequent player always having the most points. The first item always goes to the first player, the second item to the second player, and so on. If you happen to be the 5th player in line, your chances of getting anything become practically zero. And if you join a guild with such a system already in place and everybody already having accumulated lots of points, even participating in every raid will take months before you get your first item.
Such a system has one advantage, but only for the tanks: You automatically create a "main tank" with the best equipment. There are good arguments for such an arrangement, as a well equipped main tank is important and helps the whole guild. But for the other classes the same distribution will happen, you effectively create a "main druid", "main hunter", "main priest", etc., which isn't necessarily optimal for the whole raid. Especially if the guild is able to beat both MC and Onyxia, you end up with the "main" of a class replacing his tier 1 stuff with tier 2, while somebody behind in the ranking is still running around with tier 0 items.
Thus a positive sum system can create lots of problems: For example recruitment is difficult if the new recruit learns that he only gets his first epic after somebody else gets his next 10 epics. Another possible problem are guild alliances, where a non-even distribution tends to cause lots of inter-guild political strife. But probably the worst disadvantage of the system is that it creates a growing gap between the first player(s) of each class and the last players of the same class. There is a substantial risk that after you have your main tank equipped with full tier 2 gear, while the other warriors are much less well equipped, the main tank either gets fed up and quits the game, or he switches to a more uber guild, and sets back the development of the whole guild in the raid circuit by several notches. Of course people hate setbacks, which causes lots of guild drama, more people leaving and starting a death spiral of setbacks, so the main tank leaving is one of the major causes of guild death.
So personally I would prefer either a zero-sum system, or a system with bonuses and rewards where the average number of points given out during a typical raid is very, very close to the amount of points spent during each raid, thus being nearly zero sum. One way to achieve that are systems where the cost for an epic is not fixed, but where people can bid points. If people are honest, such a bidding system tends to make players spend more points when they have more of them, thus automatically correcting the positive sum imbalances. Unfortunately the disadvantage of bidding systems is that players have a tendency to collude. For example all the hunters agree not to outbid each other, but work out a separate distribution system among them, and then if the warriors and rogues don't do the same sort of collusion, the hunters end up with more points on average, and are able to outbid the other classes on multi-class items like the famouse hunter weapons. No system is perfect.