I started to look into this as well. I have a project in standby. All out in the public but quietly. So, what I did …
With just 1-2 users (myself), memory would sit at around 145 MB. Then in the image you can see (on the CPU) the time when I run a load test (with LoadImpact) with 50 users. CPU went 100% and another box was started (I use MUP with AWS EBS). Test load was finished, memory remained up (~210 MB) until my next deploy (or restart). In the second image, 10% CPU is when the app was re-deployed and the memory freed. Also after that, there are those small memory waves associated with small spikes in CPU, a pattern that I started to see with 1.9-rc.2 and was curious to know how it looks on a highly loaded server with many users. Do the waves and the spikes grow in amplitude or are they as unnoticeable like on my almost standby project.
