Doesn't really matter -- as long as the odds outages at both sites are independent, and relatively low, it won't happen at both at the same time, usually. You can make the status service pretty reliable, cheaply, compared to most other services, and if it loses the central servers, the remote monitoring nodes can still store test information, so when the service comes back up, the historical record should be accurate.

true, but let's add to the criteria:

reliable, accurate, high-frequency