I didn't buy that statement on the APC article that said they claimed without any power/cooling changes you lose efficiency. Right there in their document on page 29 it shows a big chart that without any changes to power/cooling you save 29% in their model, with optimal changes those savings go to to 65%.
If you have workloads that are compatible with virtualiation, there is no doubt that you can be more efficient with virtualization than without. Now if you implement it very poorly it may not work out, but I'm talking purely on the server/storage/networking front. Don't need to touch the cooling or power systems to see big gains in efficiency with virtualization.
Even without virtualization you can see big gains in efficiency if you optimize the power/cooling, think blades, DC-based servers, servers that run multiple systems off a single power supply, right sizing power supplies for the servers, etc. At least in my line of work, I work with co-location faciltiies, as does everyone I know, so we have really no control over the power/cooling of the facilities we're in, though we can still get more out of our hardware with virtualization than without. Means needing less cabinets, less power circuits etc.