Exactly. And when you're pumping out half a dozen titles, each with wildly different performance charactistics and no idea which will bomb and which will be a huge hit, it's not really worthwhile to be overly thorough with any single title until the shit hits the fan...
However, I do think the case could be made that trying to roll your own infrastructure in this kind of scenario is perhaps not the best use of your time, when solutions to scalabilty problems already exist and are so well tested.
(Edit: what I mean by that is, if you use something like App Engine, you really don't need to be concerned with scalability all that much -- as long as you're working sensibly and within the constraints of the platform)
However, I do think the case could be made that trying to roll your own infrastructure in this kind of scenario is perhaps not the best use of your time, when solutions to scalabilty problems already exist and are so well tested.
(Edit: what I mean by that is, if you use something like App Engine, you really don't need to be concerned with scalability all that much -- as long as you're working sensibly and within the constraints of the platform)