My impression from following open clinical trials (in order to participate) is that the bulk of antidepressant research is conducted on outpatient subjects -- people who can drag themselves to a research facility, sign up for a study, go through an intake process, and risk a 50% chance of placebo. Because few severely depressed patients can do all of that, I suspect that severe depression is underrepresented in the subject pools. No wonder, then, that the trials show low efficacy; it's well known that the drugs work best (or only work) in cases of severe depression.
The last time I was acutely depressed I called my doctor the day after Christmas to schedule an emergency appointment (it took everything I had not to call her on Christmas). If I hadn't been able to see her I would have gone to the nearest ER. There is no way that I would have had the wherewithal, the ability, or the spectacularly bad judgment to participate in a study instead of seeking immediate treatment.
Severe depression is also underrepresented in pools of research subjects because there are ethical proscriptions against allowing obviously suicidal patients to participate in studies. Those patients are referred to a hospital instead. In other words, patients have to be reasonably well in order to participate. Moreover, patients with delusions or hallucinations are often disqualified from participating in research on depression unless the study is looking specifically at psychotic depression.
These limitations may explain why antidepressant medications show more clinical benefit than the research predicts. After all, in real-life practice doctors treat patients from across the severity spectrum, whether or not they would qualify for a drug trial.