Well, a lot of people here think that their web-delivered software as a service offering is going to be an instant mega-success and they will need to scale up to meet 100k requests per second in the first day.
The truth is a much more pedestrian than that. For one thing, outages can be good publicity; I'm willing to bet many more people heard about twitter because of people bitching about it being than any other reason. For another, you should never look further than one order of magnitude ahead when scaling a web app because the load profile changes too unpredictably as you climb the curve; you can't tell what 100k visits/hr looks like from 1k visits/hr but at 1k v/h you probably have a good idea of what will kill you at 10k.
The truth is a much more pedestrian than that. For one thing, outages can be good publicity; I'm willing to bet many more people heard about twitter because of people bitching about it being than any other reason. For another, you should never look further than one order of magnitude ahead when scaling a web app because the load profile changes too unpredictably as you climb the curve; you can't tell what 100k visits/hr looks like from 1k visits/hr but at 1k v/h you probably have a good idea of what will kill you at 10k.