There are several ways to bottleneck a server. I/O subsystem bandwidth, memory bandwidth, cache hit ratio, and CPU horsepower.
While certain type of applications are CPU bound, many many more are memory or I/O bound.
A good example is the fact the PostgreSQL runs 2 to 4 times faster on a multi-way opteron than it does on a multiway xeon of the same class, due to memory bandwidth and access. Throwing more CPUs at a problem, can sometimes actually make a machine run slower due to switching contention.
That said, I think for certain kinds of applications (jboss servers, for instance) this multi-threaded sun CPU is a good idea.