I worked for one of the largest airline reservation systems. It is a very complicated space, many degrees of complexity above your run of the mill social networking website. Unfortunately, the underlying technology goes back many decades (it is mainframe based, I am not sure these other 500 servers they mentioned do). I think that with the newer tech out there, it could probably be re-engineered to be totally fault tolerant but it would be a massive undertaking in $$$$$. To give you some clarity of the complexity, the system even calculates the weight distribution of planes as passengers check in and clears them for take off accordingly.