Doubly cached data
One reason for not making the shared buffer cache too large is that the OS cache is also being used for reads and writes, and it's extremely likely that there's going to be some wasted overlap there. For example, if you read a database block from disk that's not been requested before by the server, it's first going to make its way into the OS cache, and then it will be copied into the database buffer cache, what's referred to as double buffering. The OS copy and the database cache copy will both be evicted from their respective caches eventually, but for some period of time there's going to be duplication there. Keeping shared_buffers at only a modest percentage of RAM reduces the amount of such data likely to be duplicated.
Inspecting the OS cache
On some operating systems, it's possible to extract information from the operating system's cache and combine it with details of the PostgreSQL one. If available, this provides...
