Regarding you point on peak demand, I would expect that any 'autonomous taxi service' would use demand based pricing to mitigate the impact of peak hours. Essentially this means, that for a lot of people, the cheapest way to get to and from work will still be public transportation/biking/walking (at least during peak hours). Of course, if autonomous vehicles become cheap/ubiquitous enough, at some point along the price curve, it would make sense to have a ton of extra capacity to satisfy peak demand. Also, car sharing services would add another factor into the equation. Taxi companies would need to be profitable, whereas car sharing companies would only need to return enough value to the individual car owners to convince them to participate (i.e. participating in the service does not necessarily have to be 'profitable' for the owner, it just needs to offset enough of their costs to convince them it is worth it).

On your point about sprawl, I can just as easily envision the opposite happening. As car ownership becomes less of a necessity, I do not need to worry about a garage or parking when considering where to live. Additionally, I don't think that autonomous vehicles will necessarily make car use cheap, I think more realistically, the marginal cost of driving (i.e. cost per mile) will still be a non-insignificant amount, but the convenience will increase dramatically (especially for people who do not own cars), but I imagine commuting an hour to work each way using an autonomous taxi service will still cost you a pretty penny (partly due to the demand issues mentioned above). Also, long commutes still suck, even if you aren't driving the car. I participated in carpool for a year (one hour each way, my car was too crappy, so I never drove, just pitched in for gas), and I was able to get some work or reading done, but I still hated it.