It is awful. It makes the standard way to plan and do a DNS migration really difficult.
It occurred rarely, but a few years ago it was a regular problem because some bigger ISPs were doing it[0]. Not sure how common it is these days.
That being said, I'm having troubling coming up with a project that would be better served with IP addresses than a constant name, so I have no idea what the OP I was responding to could be doing that that problem needed to be addressed at all.
Unless you have a reference for Google saying they purposely do this, it is more likely that the cache is dropping LRU entries as it fills up. Also, I doubt there is a trivial number of actual resolvers behind the google public DNS endpoints, so you may be seeing the result of multiple individual servers without shared cache initially populating their caches.
Some quick tests with dig seem to indicate that, at least for the region I'm in, my queries to google's public DNS is rotating between 4 or 5 servers, as evidenced by the TTLs being returned.
I assumed the grandparent simply didn't understand the need to lower his TTLs.