I agree with @Joey
Virtualisation is way overrated. And often costs significantly more than discrete hardware with a decrease in reliability.
Take my environment for example.
I work for a (non-UK) government department.
Our Sun servers are outsourced.
Our maintenance costs for a T5220 is something like 3k/month for the server, plus 2k/month for each solaris zone (virtual server). Therefore for a piece of kit with 3 VMs on it costs 9k/month, which is made up of 3k/month for the 'global zone' (the administration zone that only our outsourcers have access to) plus 2k/month for each of the VMs.
Previous to using VMs, it cost a flat 3k/month for each server. Therefore having 3 physical servers cost 9k/month.
Now of course, if we have physical hardware failure we have lost 3 zones, what was once 3 physical servers, for the same cost.
Whie you might say that there is more flexibility, in that there is the ability to assign unused resources from one zone to another, while in theory there is, it costs $$ to get the resources assigned. First, someone has to prove we need resources moved from zone A to zone B (say 20 man-hours), then they've got to convince their management to send a request to the outsourcer to move the resources. Then it's got to go through change control (another 5-10 man-hours of various people's time), then the outsourcer, 4 weeks later, charges 4k+ to type a few commands on the command line to move the resources...
From an administrators point of view, manageing 10 VMs is exactly the same as managing 10 physical servers that aren't VMed. It still takes the same amount of time to ssh to a VM, check logs/restart an app server/truss processes/configure something/deploy something as it does to ssh to a physical server. I can definitely tell you that labour costs of system admins managing a server, whether it is physical or virtual, are far greater than labour costs in asset management or data centre costs. And with the virtualisatoin hype, management (on advice of the outsourcers of course) are going to create shedloads of VMs, costing more $$$ than having a few physical servers.
Using VMs also increases complexity and other maintenance costs. Does this new patch work in a VM? Do we have to keep VMs patched? How much does it cost to verify the software/patch that works fine on a bare-metal server also works within a VM on the same hardware?
What about tracking down problems? What is the real physical interface traffic is coming into/out of because in the VM we only have visibility to the virtual interface? How does it complicate our routing tables on the server to ensure that data leaving VM A has a source IP address of A while traffic leaving VM B has source of B, rather than each one randomly picking a physical interface to transmit out from?
Virtualisation just increases costs and decreases reliability.