When I started to do open source development, I tried to speak at several OS conferences. In most cases I didn't hear back, but once they told me in no uncertain terms that they didn't tolerate "no-name" speakers (even though the signup page implied everyone was welcome).
The other way would be if you personally know some movers and shakers. I think that's also quite common. From time to time it's so obvious you're going to see speakers talking about projects or subjects that couldn't possibly have gotten up there on their own merits. They're not terrible, but they're sure not doing stuff that's significantly cooler than what most other people in the community do.
The third option is that you're so well known in your community that you technically don't need corporate sponsorship or personal connections, but those probably go together anyway more often than not.
Everybody else is going to have a bad time. Im uncertain if this is a bad thing per se, but it's also easy to see how that precludes the participation of a lot of people who could make a nice contribution. But if speaking is what you absolutely want to do, you could conceivably work your way up from smaller to ever-bigger conferences.
For 2014 we're more than likely going back to 95 slots, and I'd expect the number of proposals submitted to be roughly the same or higher. That'll put us closer to 20%.
Plenty of mere mortals get on stage at PyCon. Speaking experience does hold some weight with reviewers, but a well organized and interesting topic will take you pretty far.
We always schedule at least some first time speakers, because, well, that's how you get your next batch of experienced speakers.
On top of this, anybody whose talk isn't scheduled tends to be encouraged to give a lightning talk (5 minutes), and anybody who gave a particularly memorable lightning talk one year is almost certainly going to be accepted the following year.
I would expect that a lot of non-profit community conferences end up along these lines; the expensive commercial ones much less so.