For those power users who can't (or won't) "share" a server, there's a different flavor of VDI available. You still put the client machine in the data center, but dedicate a blade server to it. A significant investment for hardware dedicated to a specific user, but the design of blade server systems still reduces overall investment. While you scale the amount of memory linearly for each dedicated blade, economies of scale are still available for resources like power supplies, network connections, and cooling.

Of course, all three options still require some kind of interface equipment at the end user location—after all, there has to be a screen to look at and a keyboard to type on. For the first option, the current hardware can be left in place. For the latter two options, an existing desktop can be use. However, a thin client is also a possibility. This presents the intriguing opportunity to implement VDI for current users, using their existing desktop machines; when new users join the company (or a desktop machine for an existing user needs to be replaced) a thin client device is given to the user. The cost differential between a fully-scaled desktop device and a thin client can be very large; I've heard $200, quantity of one, quoted for thin clients.

Of course, there are other factors to be considered. The capacity of the network needs to be examined to see if it can handle the traffic between the data center and the desktop locations. Also, the cooling capacity dispersed to end user locations must be available within the data center.

On the other hand, client device reliability should go up significantly with centralized administration. Operations staff on-site visits to end user device locations should drop as a result.

I don't know that I've seen any good case studies on how the numbers might pan out for a real-world desktop virtualization implementation. Intuitively, however, desktop virtualization seems like it must be less expensive to run. However, it's often the case that a situation that, early on, made economic sense, when seen through to a later phase, doesn't make sense at all. Inertia keeps the no-longer-making-sense system in place because "it's the way it's always been done." Desktop computing made a ton of sense early on when it delivered computing ability to end users at a fraction of the cost of mainframes. It makes sense as a way to let end users perform their own computing taks. It certainly has fostered innovation because many, many apps make sense only in an end-user computing environment (the Web, anyone). But that doesn't mean that the only way to achieve those benefits in the future is to plunk an expensive, extremely powerful computing device in front of the person interacting with it.