I suspect the biggest, but not the only, problem with these supposedly weak SSRI/SNRI numbers we routinely see in clinical trials has to do with the definition of depression, with the diagnosis itself. For example it has some of the worst inter-rater reliability across the entire DSM (meaning that two doctors are least likely to both reach the same diagnosis in the same patient independently). So if you start from a poorly defined set, which likely encompasses some genuine affective disorders, people going through difficult times, undiagnosed personality disorders, dysthymic ADHDers, burned out ASDs, and God knows who else - yeah, you'll get poor performance data. Every psychiatrist knows intuitively that SSRIs/SNRIs do work, even if you have to trial a few. Also, without arguing about the number, that 15% is not exactly the same 15% for each drug.

Also, since we're here: the secret knowledge about depression (and affective disorders in general) is that it is an episodic illness, where episodes are measured in weeks or in months. Most people experience just one episode which will end whether or not any medical intervention is undertaken, although the intervention can greatly shorten the course and avoid a potential suicide. But some will not stop at one episode, and can go on to have multiple episodes, perhaps develop melancholic features or even suffer from a lifelong recurrence. It is in the latter groups where medication is the most effective.