1. Much higher occupancy per car. Based on my personal observations, at rush hour the average occupancy of a car is something like 1.3. I don't see any reason that can't increase to 2.6 or more without significantly slowing anyone down -- 2x
2. Yes, a lot of people need to get to work at 8, but normal working hours start at different times for different people, anywhere from 6:30-9:30 is fairly typical in a U.S. city. If the average person has a 30-40 minute commute, and there is a reasonable amount of reverse commute demand, a car should be able to make 2-3 round trips per day at least -- 2.5x
3. Cars driving more efficiently with quicker reaction time and less mistakes will speed up commutes a lot. Hard to say how much but I'd say at least 30%. This directly increases capacity, since shorter commutes mean each car moves more people in the 6:30-9:30 window -- 1.3x
4. If self driving cars can communicate with each other to park, they can pack themselves very closely, and make room for a car to get in and out, removing the need for oversized parking spots, space between cars for doors to open, and aisles between cars. This should improve parking density by at least 3x.
5. Cars that are on-demand and self-driving can be smaller than typical cars we drive today. Since you can order any sized vehicle based on what's needed for an individual trip, there will be many fewer empty seats. Without the need for a steering wheel and controls, interiors can make much greater use of available interior space. This should reduce the exterior size of the average car at least 1.5x.
6. Cars can find parking in more remote areas. If parking is not available within a block of the office, the car can drop you off and park a few blocks away. This increases utilitzation, say 1.2x.
So, multiply it all together, and you get 35x better parking efficiency. It's going to be a huge win for dense cities.