Measuring the Memory Overhead of a Postgres Connection

One fairly common complaint about postgres is that is that each connection uses too much memory. Often made when comparing postgres' connection model to one where each connection is assigned a dedicated thread, instead of the current model where each connection has a dedicated process. To be clear: This is a worthwhile discussion to have. And there are several important improvements we could make to reduce memory usage. That said, I think one common cause of these concerns is that the easy ways to measure the memory usage of a postgres backend, like top and ps, are quite misleading.