1. The immediate costs of war are clearly awful. Most wars lead to massive loss of life and wealth on at least one side. If you use a standard value of life of $5M, every 200,000 deaths is equivalent to a trillion dollars of damage.
2. The long-run benefits of war are highly uncertain. Some wars - most obviously the Napoleonic Wars and World War II - at least arguably deserve credit for decades of subsequent peace. But many other wars - like the French Revolution and World War I - just sowed the seeds for new and greater horrors. You could say, "Fine, let's only fight wars with big long-run benefits." In practice, however, it's very difficult to predict a war's long-run consequences. One of the great lessons of Tetlock's Expert Political Judgment is that foreign policy experts are much more certain of their predictions than they have any right to be.
3. For a war to be morally justified, its long-run benefits have to be substantially larger than its short-run costs. I call this "the principle of mild deontology." Almost everyone thinks it's wrong to murder a random person and use his organs to save the lives of five other people. For a war to be morally justified, then, its (innocent lives saved/innocent lives lost) ratio would have to exceed 5:1. (I personally think that a much higher ratio is morally required, but I don't need that assumption to make my case).
Are there conceivable circumstances under which I'd break my pacifist principles? Yes; as I explained in my debate with Robin Hanson, I oppose "one-sentence moral theories"