Facebook are definitely not relying on the database, they virtually have everything in cache. You can't scale/distribute an RDBMS the way they are growing- more than half a million new users every day. And it's not only them.
If there are problems, they are not with Scala, as some of the highest-traffic sites use it, for instance Twitter. It's more likely that it's a problem of ESME or maybe Lift, as we are using the SNAPSHOT branch and they had many changes lately. But I definitely remember that ESME was faster one or two months ago. Maybe we also have to try to profile the stack trace where the biggest number of String objects get allocated. It might be slow, but will certainly clarify many questions. The important thing to do it is that we are easily able to reproduce it, so selenium/load tests would be cool.
