Price alone is not necessarily indicative of performance, especially when you're comparing products across decades. In the days of aviation INS units, computers used to cost a lot too. Advances in silicon fabrication have changed all of that.
While the sensors onboard a drone aren't the ultimate in terms of accuracy, they're enough to get it at least heading in the right direction. And I still think that improves the odds and is better than nothing. It's easy enough to limit the non-GPS RTH flight time in order to prevent the drone from going too far afield.
I'm speaking in terms of principles here. But it would be really interesting to see an example of non-GPS dead-reckoning navigation to get an idea of just what it would or wouldn't be capable of. You'd think that a sufficiently savvy person would be able to cook something like a phone app up as a demonstration.