The buses need to run at night so the people can rely on them during the day -- it's just the price of public transit, the public uses transit way more if buses run on 10-minute interval during off-peak rather than 30-minut interval. (another study I can't find right now).
I think as programmers we tend to try and optimize problem for least pollution, but the problem here lies in the domains of sociology and urban planning; we might not be the most qualified people to solve it.
For example if buses pollute 3% because of the extra buses on off-peak routes but now more people use them, this would have net effect of reducing pollution.
All your points are correct, but you're not responding to the point that SDCs can be superior to buses for those low-capacity times. If it were one or the other, then yes, buses could be less pollution on average, but that doesn't address the case the parent was talking about.
I've been on double-length public buses where I was the only passenger for most of the journey. And many times where there were only 1 or 2 other people. This is necessary e.g. at night where a bus only runs every 30 minutes, there has to be a bus, but there just aren't that many people.
Obviously in this case individual cars would pollute much less, and even take up less space.
Any realistic solution for a city is going to combine self-driving cars for the long tail of origins/destinations and times, with public transit for the busiest routes at the busiest times, including links between the two.
I'd love to be able to take a self-driving car from a house in Queens directly to the subway, the subway to Brooklyn, and then another self-driving car to my final destination that's 20 blocks away from the subway.