Optimizing stuff like this is fun. If I were a betting man, I'd put money on the line-of-sight calculations as the guilty party for large CPU loads. That area could probably use optimization, but ultimately there comes a point where the code can't be optimized further and the load should be handled by > 1 server.
Maybe the people who run guildwars can set up a profiler of some sort to determine what causes the load. The quickest and easiest profiling is to see which thread is getting the most attention. There's only a limited number of threads in the app, and all of the line-of-sight calcs are done on one thread.
For APPLAYER in particular, my guess would be to make the object smarter and bear the memory alloc. Most instances of the object end up being copied several times before making it to the network pipe. Doodman had a good idea along these lines a while back, maybe he could share it here
Threaded Mode
