There is an article on Network World which is titled "Virtualization Projects fail to reach ROI targets". The main reason provided was that customers believed too much into hypervisor vendor provided models which are skewed to illustrate ROI. I agree with the author that there is some truth to it.
I started to look at major vendors like VMWare, Cisco, HP, Microsoft, IBM, etc and their version of Virtualization ROI. They are talking mainly about energy (power & cooling) savings - which can be easily demonstrated as $ spent/saved. What these don't take into account are the other costs to make this saving happen.
Here is an example: Datacenter wants to move 50 web servers to virtual environments. If they just move 50 to such an environment + add a few more to compensate against performance, then ROI can be achieved. But, if they now state, I want to move my switch to a virtual switch, my firewall to virtual firewall, my Single Sign On solution to vitrual platform, you are asking for trouble in terms of ROI as these were never accounted for (professional services costs, time to deploy, cost of new licenses, management infrastructure, security/threat modeling, etc) in the first place. There is nothing wrong in doing it, it just needs to be accounted for.