Citation needed.
Even our Active-Active cluster boxes have redundant power supplies plugged into seperate electrical circuits and wired to independant UPSes, never mind our Active-Passive cluster solutions...
The fact is, most data centers do go for maximum redundancies without single points of failure on the hardware side.
When you have a massively parallele solution with custom software that is built to run on non-redundant hardware like Google built with their search engine, yeah, you can afford to skimp on hardware. They don't care if 1 node out of their 10000 fails, and the software doesn't see the impact. But that 1 specialised custom application is not an industry standard and is far from the norm in building data centers.
The fact is, the Xserve wasn't selling well and it had all the server features. A rackable Mac Pro would sell even less to those Xserve buyers. Forget redundant power supplies if you don't believe in them, just lack of LOM or hot-swap drives is a killer by itself.
And seriously, Thunderbolt ? Host based storage ? Forget that, to get into my data center, you need multi-path Fiber Channel. Thank god at least Apple recognizes that and offers the option on the Mac Pro. Thunderbolt is not a SAN technology and it's not replacing SANs anytime soon. I don't want to manage hundreds of storage arrays for each hosts. I want to manage 1 unified storage array and then present LUNs to my hosts as needed. That way, I get better distribution of my existing storage and can even manage some over-provisionning depending on the technology I use.
A lot of people here never worked with enterprise-grade systems. A rackable Mac Pro would at best be used as someone else stated, to rack along video/audio equipement in a studio. Not to rack into a data center.