We're putting a bunch of fun new code live, which eliminates a ton of global database queries. The global cluster is the machines not associated with a single user, but instead all of them.... like the global user list, friends data, user update times, etc. That cluster's been growing steadily more busy over time, but we haven't been buying new machines for it, because we knew they wouldn't be necessary. Tonight's update will reduce the global query load to almost nothing, putting it all in memcached.
The other thing we'll be doing is moving a bunch of users off the Ribeye cluster and onto another empty database. (We often shuffle users around clusters to balance load and performance hardware/software upgrades on machines in the cluster.) Ribeye's been extremely loaded lately, so moving users off of it will help a lot. The move won't finish tonight... it takes a couple days (5-10 users at a time...), but the site will speed up as it goes. We'll be moving active users first.