Large scale problems are really optimization problems combined with distributed systems and large app (multiple team member) complexity. I'd argue that until your userbase is large enough to need a distributed system you probably don't need to change much between large scale and small scale. Good optimized code is good optimized code.
That said, it definitely helps to read up on large scale distributed systems to see how they do things. That will help you understand what complexities might occur if your system eventually scales.
The largest scale is global. Chances are if you have that kind of scale then you have bucketloads of money and manpower to throw at the problem. At that scale timing of multiple systems becomes an issue with server performance. There's a big difference between distributed systems in a local market vs one that needs to synchronize performance with multiple localization issues like language, legalities, sociological concepts, interface changes, etc.