Timekeeping of atomic clocks is still waaaayy better than next-best technologies (like temp-compensated Xtal oscillators). And size/power/$ constraints matter. So CSACs for the win imho.
There is also, of course, the issue of infrastructure dependence. Particularly since wireless telephony has moved almost exclusively to GNSS time we're going to have a really bad time if kessler syndrome takes out the GNSS satellites.
Edit: Here is an example adev chart for a inexpensive atomic clock vs what appears to be a pretty good timing GPS receiver: https://www.thinksrs.com/images/instr/prs10/PRS10diag2LG.gif
So in that case the GPS accuracy only beats out the free running atomic clock at intervals greater than 200,000 seconds or so.
Here is a collection of older timing receivers: http://www.leapsecond.com/pages/3gps/gps-adev-mdev.gif
A quick look didn't turn up any cheap non-timing receivers, but my experience is they're pretty bad (I mean relative to atomic standards, of course).
There are better timing receivers than the ones charted above, of course, but they are not $5. Their cost is now in the same general ballpark as surplus atomic clocks.
Of course, if you have both you can sync one to the other with whatever time constant maximizes the composite performance and have the best of both.
(and primary atomic clocks don't have this drift issue, but sadly the days of the occasional sub $1000 5071 showing up on auction sites seem to be over. :P )