I think this is a damned-if-you-do, damned-if-you-don't situation though. Scalability doesn't come for free, in a small startup it usually costs you velocity (hence: "Do things that don't scale"), and with Facebook breathing down their necks that may not be a luxury they could afford.
But since video chat is quite commoditized building a scalable system from ground up would have helped them counter facebook by quickly building features on top of what they had. The article itself mentions that they had to stop what they were doing and scale the system.
My main point was that one-size-fits-all methodologies don't suit every software project. In some cases it makes sense to build the feature first and then worry about scaling but in some cases like this it doesn't.
