From an interview with Sun CIO Bill Vass... I'm quoting an interesting part on how he differentiates between different server architectures, and the best application for them:

Because each server is capable of running the same applications, it can be confusing how to choose one over the other. Since I really like cars, I think of the servers in terms of capacity and speed. For mass transit-- such as an application, portal, Web, directory or proxy server--you might need as many as 32 threads at once. That's what the CoolThreads server offers. It is like a bus.

When you need to share memory quickly, in a large database or an ERP [Enterprise Resource Planning] system, you'd use an UltraSparc. There aren't as many threads, but it shares memory well from chip to chip, and it is pretty fast. I think of it like a BMW M5 sedan that can carry a whole family, with room for luggage.

Finally, the x64 Opteron--AMD's chip--runs very fast on two- and four-threaded applications. That's the two-seater Ferrari.

People are used to the idea of a general-purpose chip to do everything in the IT space. We're working on an application profiler where CIOs can enter information and generate a recommendation for how much memory-sharing they need and which chip is best for any given application.