There are a handful of mathematical theorems that basically say "there is no perfect voting system." More rigorously, they enumerate a list of desirable properties that you would like voting systems to have, and show that NO voting system can have all of these properties, because they are actually contradictory.
Some of these properties are things that are "obvious" ones that any practical voting system should satisfy: voting matters, there is no dictator, candidates can win, determinism. Two of them are non-trivial: absence of tactical voting, and irrelevance of independent alternatives (IIA). Tactical voting basically means you have a game-theoretic incentive to vote in a way that differs from your true desires. IIA basically means that if Gore would win a 2-man election against Bush, adding Nader to the race should not make Bush the winner. These properties are related, but not the same.
All of these properties are desirable, and a "perfect" voting system would have all of them, but you can't have all of them. You have to pick at least one to give up. Tactical voting in particular is extremely hard to remove, even if you give up IIA, while systems that satisfy IIA are somewhat complicated. The article suggests that it is very easy to satisfy all of the other criteria, if we let go of one of the "obvious" criteria: determinism.
So it's basically a question of picking your poison. You can have determinism, or you can get rid of tactical voting, but you can't do both. The author is contending that determinism is the best property to give up.
Further reading: http://en.wikipedia.org/wiki/Arrows_theorem is one of the mentioned theorems, dealing with ranked-preference voting systems (vote-for-one-candidate is a special case). There are a few others for different classes of voting systems.