My expectation would be that all boot jobs that depend on an event that is guaranteed to take place after the reboot invocation returns won't actually get started.

Experiments prove that this is not the case: If you put one (task) job that does a reboot, and then have a second job depending on the first append a line to a file, the line will end up in the file.

Given multiple people are surprised by this behavior and the risk of it causing subtle issues due to reboot taking place at the "wrong" time, we should look into upstart logic to see whether this is intended behavior or should be fixed.