Wiki defines Desktop Virtualization (DV) as "Desktop Virtualization involves separating the physical location where the PC desktop resides from where the user is accessing the PC." They go on to define the following flavours :

"Single Remote Desktop - in this model, a single desktop PC is accessed remotely across a network connection using remote PC access software, such as GoToMyPC, WebEx, PCAnywhere, Windows Remote Desktop, VNC and other similar methods.

Shared Desktops - in this model, a multi-user server PC environment like VDIworks Virtual Desktop Platform, Quest's Provision Networks, Citrix, Ericom Software, WebGlobix Software, and Terminal Services are used to host many users who all "share" a common PC desktop environment together on a server machine. In this case, it's generally possible to host up to a few hundred desktop sessions on powerful server hardware.

Virtual Machine Desktops - in this model, also referred to as Virtual Desktop Infrastructure, or VDI, virtual machine technology is used to host multiple instances of a standard, single-user desktop PC operating system (e.g., Windows XP) on a server machine. In this case, it's generally possibly to host a few dozen desktop sessions on powerful server hardware. VDI products are offered by Quest's Provision Networks, VDIworks, Citrix, VMware, WebGlobix and Ericom.

Physical PC Blade Desktops - in this model, individual "client blade" PCs are used to host multiple independent user sessions, each one running on its own physical PC blade. In this case, it's possible to host as many client PC blades as you have rack space, power and data center space to accommodate. There are several companies that provide this model such as HP and ClearCube Technology. "

Bill Snyder makes a case for an overstated business case largely on account of unclear ROI drivers. Jonathan at Information week takes a contrary view, arguing that DV drives security besides just dollar savings. He brings out the following benefits. My thoughts are in italics:
- Stricter licensing. Sure benefit here. I am sure most managements would not like to be burdened by the consequences of employees installing "evaluation" versions and failing to remove them. I also believe there are other ways of controlling this centrally without going the DV route.
- Better manageability.
- Lower TCO on account of extended thin-client hardware life, fewer cycles spent on hardware-induced OS failure, and lightened deployment efforts. Extend client hardware life and lower OS failure costs ? Not sure if there is a real TCO case here. Clients are usually replaced not because the hardware has failed, but because the s/w has caught up and is more hungry. You would need to upgrade the server infrastructure if this happened in a DV environment and the exact cost -benefit comparision is not clear. But yes, if you are not running anything on the client, you would save dollars on upgrades. Hardware induced OS failures are not common enough IMO and with the typically good quality h/w found in most organizations, this is less of an issue. 
- Better business continuity. Maybe ? Depends on whether your continuity efforts are helped by letting users working offline rather than depend entirely on a centralized environment where they need to be online all the time to be productive. In countries where the physical infrastructure (roads, transportation etc) is strained there is a trend of letting employees work from home. In such cases, if you also had sub optimal telecom infrastructure, you might get better continuity by not centralizing. 

Guess the jury is still out on this one.