>> I believe that they have about 6000 Linux servers now. That's what I
>> call a reasonable sized server farm!
>
> And I'll bet they could have done it more cheaply with one mainframe.
> LOL.

Not if the tasks are compute intensive, which I suspect they are.
Modern mainframes have only slightly more CPU cycles than high-end
PC-class servers.

Yes, you *can* run thousands of partitions on a z-series or system/390
machine, but they don't get very many compute cycles each.

Those guys at Google are pretty sharp.  If a mainframe would be a more
cost-effective solution, they'd probably be doing that.