More granular configuration of webserver worker types, proxy buffering, etc.
Currently all API endpoints are served by one gunicorn server. The upload and (raw) download functionality forces us to use async worker (gthread). This also forces us to use no buffering on the proxies. This is all good and dandy for upload/download, but most request would be better served with simply sync worker and buffering.
We should think about breaking the api into two different one, served by different gunicorns or a different WSGI.