Re Low Drift Clock
Purely as a thought experiment, I wonder if a ground based clock could be used. Broadcasting a synchronised signal regularly these could be picked up and echoed back to the transmitter. The lag would give an indication of the straight line distance plus an internal delay time (signal processing etc.)
On it's own this would probably not provide sufficient information for the ground based system to provide clock rectification but if each recipient also broadcast to it's neighbours (as must happen already to allow chaining of the signal beyond the visual horizon) then they too could rebroadcast that signal with an individual identifier.
Analysing these signals would allow the creation of an exact image of the net with different time delays due to the differing distances. It would now be possible to transmit individual time sequences such that each receiver would receive the time signal simultaneously (OK - general relativity sort of negates that but it is also calculable if necessary) so that the clocks on the satellites could be reset every whatever time interval to ensure they were all withing 40 ns of a specified target time.
Slightly complex but should be doable and retaining the time source on the ground would make the individual satellites a lot cheaper - multiple ground stations would also obviate a single point of failure.
Biting the hand that feeds IT
