I am, apparently, off by an order of magnitude. The streaming API gives you more like a million tweets per day. So this size of corpus would take you a couple

I am, apparently, off by an order of magnitude. The streaming API gives you more like a million tweets per day. So this size of corpus would take you a couple of weeks to acquire if you watch the stream yourself. Which is still apparently not much longer than the java downloader takes.