Gary, I'm not necessarily disagreeing, but stating that as a DBA, I don't like to put enterprise level tuning of OLTP and OLAP SQL servers in the hands of non-DBAs. I think all technologies can be "tuned" to their maximum potential, my point is that in practice, that requires a level of expertise and integration between server and DBA personnel that most organizations don't really have.

For example, given an equal choice between a midlevel server, one could VM it or multi-instance it - give the same resources to each. Implement database mirroring as a hot failover solution, get the benefits of transactional integrity compared to hardware level solutions that may not respect transactions. You could cluster either scenario. The performance requirement is that a failover must perform equally to the primary server. I wouldn't want either the VM or multi-instanced machine to share with non-SQL servers. Given that, the additional overhead of VM would theoretically reduce the maximum potential of the machine. With hot swap memory, disks or hba (depending on which you are using), the same machine can expand without downtime.

You are correct that data centers have too many under-utilized machines, and if the DBAs are not using resources well, that is a good motivation to consolidate using VM. The right tool for the right job most certainly applies to which strategy works best for a company, as does understanding whether the infrastructure or DBA personnel are in the best position to maximize resource use of the database servers.

Still, for performance/DR/failover scenarios running SQL2008 EE in heavy OLTP and OLAP, I don't think even the best tuned VM can deliver performance at a lower cost than the best tuned physical, multi-instanced machines. I'd love to be wrong, as I can see the potential for multi-instancing on VMs if the numbers add up!!