We run servers as virtual machines primarily for a few reasons
--consolidation
--lower costs
--reduce power consumption
--reduce cooling needs.
Generally speaking, in the past when a server was purchased, it was always purchased with more CPU power than it needed, it usually had more RAM than in needed and often had additional disk space allocated for future growth. This usually meant that the box usually sat there humming away at about 5% of it's total capacity and wasn't fully utilizing the hardware.
With virtual machines, you are able to take a number of virtual machines and run them all on one piece of hardware and fully utilize the system. It's better to run 1 server at 80% capacity than it is to run 8 servers at 10% capacity in terms of cost, heat, and cooling.
With virtual machines, in an enterprise environment, you have so much flexibility. For example, we run vmware vSphere in our environment and we have a storage area network. Thus, we might have 5 physical servers each connected via fibre channel cards to a storage area network storage device. So, we might run 50-75 virtual machines across these 5 physical hosts. So, lets say a host needs an upgrade (maybe more RAM or new processors). In a virtual environment like VMWare, you can move the VM's that are running on it to another remaining machine (while it's running with no downtime to the customer). You can then put the 1st host into maintenance mode and you can upgrade it, patch it, perform repairs and when ready you can bring it back online and move servers back to it. The key is that all of this can be done without EVER shutting down the virtual machine serving customers or employees. Of course, this does require a storage area network, which can be a $500,000 storage array, or as simple as a free Linux distro like OpenFiler or FreeNAS. You simply need shared storage.
In addition, virtual machines make OS patching a breeze. Before a big patch, you simply click a button and within about 15 seconds, you have a snapshot of the machine. Let's say you load a service pack and then all of a sudden your server just keeps blue screening and doesn't run properly. In a virtualized world, you just restore back to that snapshot and VIOLA..you are shot right back to where you were. No fuss, no muss.
In a virtualized world, lets say that you have a machine up and running...but would like to build up a test lab to test a new version of software. Simple, you just clone the original box and you have an exact point it time copy. You can bring that up on a test network and test all of you want. Once you are comfortable, you can quicky and easily roll it out to production.
In a virtualized world., you can quickly and easily add more storage space to a server. Oh, 40GB wasn't enough, give it 80, resize the virtual drive and VIOLA..you now have 80GB.
In a virtualized world, you can reboot a VM in about 15 seconds. On physical enterprise class servers, the POST process alone can take 2-3 minutes to intialize a bunch of RAM, initialize a SCSI controller, bring the RAID system online, etc. You don't have anything like this to wait for in a virtual world.
In a virtual world, with all of the hardware virtualized...you are not dependent whatsoever on the machine that you run it on. So, in the event of a crash and you have to restore, you don't need to get the same hardware, same RAID controller, etc. Since all of that hardware is provided in a virtualized form, you simply restore your VM onto any remaining VMWare (or other brand VM Server)...and its back up and running. It's totally hardware agnostic.
I haven't installed anything, but vitrualization software, on a physical server in years. The virtual environment is simply faster, easier, more robust, more powerful, more cost effective, quicker to deploy, etc. I can deploy a new windows server in our environment in approx 6 minutes. It's getting rare these days to run any type of real server on iron ( a real physical server).
