Sorry to break it to
you, but scaling isn't hard anymore. The tools we have at our disposal today makes it trivial compared to companies attempting to do the same thing 5-6 years ago. As soon as you figure out the scaling from 1 -> 2 servers, you're basically done until you hit top-1000 traffic. Because as soon as you spend a weekend doing that, scaling to 2 -> N is child's play.
Oh look, I'll pick from any number of battle-tested, high performance, and free load balancers. I'll spin up a couple more VPS nodes. I'll add a few more slaves to my DB setup. I'll install some memcached and a quick write-through wrapper in my code. Oh yeah, duh, don't do long-running operations on the request/response cycle, here's an open source queue system for you. And since I like going overboard, let's throw some Varnish all over the place. Congrats, we've just covered the scaling concerns of 95% of web apps.
Try doing that several years ago with the terrible PHP and Java services we were writing, and without most of the hardened tools we now get from thirty seconds of Googling and learning from everyone else's mistakes.
Disclaimer: yes, when you start to break through the top-1000 some hard problems start appearing again. You begin to run into the limits of what others have created, and patching / rolling your own solutions becomes commonplace.