It's important to look at the FDA's job in terms of diminishing returns, or maybe as a classification problem. You have some number of drugs coming in, some of which are helpful and some of which are harmful. But you don't have any perfect means of figuring out whether a drug is good or bad. If you start from no screening it's very easy to stop a lot of bad drugs without stopping many good drugs, but as you apply tighter and tighter criteria the ration of good drugs you accidentally stop for each bad drug you stop gets higher and higher.[1]
In practice, of course, some of the rejected drugs aren't explicitly rejected by the FDA but rather rejected by the system because they are never put through the FDA screening in the first place, but the principle is the same.
One would ideally want to set the difficulty of getting through FDA testing to maximize some quantity, such as the average effect of an approved drug on people's Quality Adjusted Life Years. But it's clear that this is not actually what happens. When the FDA rejects a drug people never hear about the lives that were not saved by it, but if even a single person is killed by a drug that was let through FDA screening. Every study on the matter that I've seen says that the FDA is actually way too restrictive in how it regulates drugs. And I should mention that this is in contrast to other countries such as most of Europe or Canada or Japan which have much saner drug regulation regimes.
I should also say that this is only a criticism of the FDA in it's role as a drug regulator. It seems to do a pretty good job as a food regulator. I suppose the number of voters involved in food production dwarfs those involved in drug production.
[1] You get more type 1 errors for each type 2 error.
http://en.wikipedia.org/wiki/Type_I_and_type_II_errors