Re: Could be why google runs on ~ 15,000 PCs. One fails they don't bother to replace it. Not 'cost effective'. It's the trend. Buy the cheapest *STUFF* that'll run the software, put in redundancy and walk away. When 1/2 the farm gets obsolete scrape it out with a front loader and start over.
It's one thing if the system is just answering search requests. There is nothing being written, and people don't care if their search takes double or triple or 10x the time (from 0.1 seconds to 1.0 seconds), or if their search returns an incorrect result from time to time. They will care, though, if important information is returned. If the information financial, medical, or personal, then an error would be disastrous. How would you like your $50,000.00 bank account reduced to $10,000.00? How would you like the doctor to give you the wrong pills and make you even more sick? How would you like looking up an address and finding directions that take you in the wrong direction?
Some computer systems require ultimate protection against faults. Redundancy does not help when a transient soft error corrupts a piece of data. RAS is still important, and to businesses, uptime is critical. Different applications require different solutions. The Google model does not apply everywhere, or even in most cases.
Market Data 
Markets 
AI
