On: NYTimes uses Hadoop, S3, EC2, and some custom code to handle PDF generation for 4TB worth of data

Take-home message: "Amazon is so fast and cheap that I can run computations over 4TB data sets, notice that I screwed it all up, and run it again."

I think what is equally important was their extensive use of open source packages of all types using both Java and Python. Xen, Hadoop, and Boto just are examples showing that more often than not, open source is the way to go.