More so, if devs spent as much time tuning the performance of their apps as they did fantasizing about "web scale" architectural pivots they would typically be farther ahead. StackOverflow.com is a perfect example of this. They run on tiny handful of windows machines, support gobs and gobs of traffic, and have absolutely fantastic performance. And as much of that is due to paying attention to performance and making sure to find and remove the bottlenecks where they exist as it is to using cutting-edge architectures like database sharding, map+reduce, eventual consistency models, etc.
Also, don't underestimate how large log files can grow in a data-driven business (like AirBNB seems to be). I could easily believe that they have many terabytes of data just from logging actions their customers have taken.
Logs don't have remotely the same access requirements as the databases used to serve a product.
That said, the blog post talks about enormous growth and it still fits inside Moore's Law's growth. I guess my gut is just saying it's not really that enormous in terms of startup scaling if it's still within those limits. Not to take anything away from 37Signal's success, but it feels like nothing of value was really added by this post. I present the post of a picture of 864GB of ram as supplementary evidence that is near the top of HN right now.
I mean, go through a list of YC companies or other startups you respect, winnow it down to the ones that exited or otherwise achieved some level of success, and play guess-the-size. How many terabytes of storage do you think e.g. Airbnb needs?