At around 11pm this evening, I will be rebooting some machines, some virtual private servers and some shell servers, to enable shared memory in all of them.
What this does is that a kernel routine scans the machines, and if it finds two identical pages of memory, it maps them both into one physical page, then anytime something tries to update one of those pages it uses a copy-on-write operation to create a separate modified copy.
This allows identical pages to share one physical page freeing up more memory for things like disk buffering and caching resulting in better overall efficiency. There are a lot of daemons that sit and run in the background from start-up to shutdown that share 99% of their memory information in common. By allowing them to physically share them, overall memory usage is lowered.