Modify the frontend code to allow the existence of multiple web servers running in a load balanced fashion. The holdup for this has been due to a semaphore lock around the code that allocates a machine for a reservation. Without the semaphore, two people hitting the site at (nearly) the exact same time can end up being assigned the same machine.
An idea I came up with to handle this is to only run the scheduling portion of the code via the XML RPC API. A single web server would be elected as the master. All web servers would call the scheduling part of the code via the API on the master, allowing the semaphore lock to still function. If the master server goes down, an election process would determine the new master.