You have to remember that Dave's speaking of a job he had in the past. I'm not sure how long ago it was, but I'd guess sometime around 2000 +/- 4 years or so, and that changes available technology somewhat. You can be relatively sure that at the time this was likely the most cost effective method of getting things done. You can be sure that there was definitely at least one microcontroller in there, but probably not a modern, power hungry, peripheral studded one with networking capabilities.
NTP's performance is heavily network dependent and difficult to make more accurate than about 1ms (I think the best examples are around 1/10th that), so it's a nightmare to use to discipline an oscillator. It takes known good absolute time and propagates it through a packet switched network, attempting to correct latency via several methods. Dave quotes the application as requiring a stability of 10
-8, or roughly 10ns (100MHz). Imagine adding a +/- 1ms jitter, that's worsening things by a factor of 10,000!
Thinking about how it might be done differently nowadays, it's hard to say. The best idea is likely to be simply take advantage of the newer technology to do the same thing. If it's within the power budget, and there's a solution to inter-device and surface communication, they might use a GPS disciplined clock source on a buoy operating primarily from solar power. It might be more cost and complication than it's worth in the long run though.
Here's
a great article from QST on building your own GPS disciplined frequency standard. It's on my list of future projects when money gets better. You can bet that Dave and the folks he worked with knew of things like this at the time (it might be related to how they synced the devices before deployment) and that they'd have used something like this if it were possible/cost effective.
Hope that helps.