As stated before, the basic idea of PgBouncer is to save connection-related costs. When a user creates a new database connection, it usually means burning a couple of hundred kilobytes of memory. This consists of approximately 20 KB of shared memory and the amount of memory used by the process serving the connection itself. While the memory consumption itself might not be a problem, the actual creation process of the connection can be comparatively time consuming. What does "time consuming" mean? Well, if you create a connection and use it, you might not even notice the time PostgreSQL needs to fork a connection. But let's take into account what a typical website does. It opens a connection, fires a handful of simple statements, and disconnects. Even though creating a connection can be barely noticed, it is still a fair amount of work compared to all the rest. How long can looking up a handful of phone numbers or some other trivial information...