Honestly, aside from learning to express a few extremely specific patterns in your language of choice concisely and elegantly and reminding yourself of the existence of certain libraries and utility functions so you don't accidentally waste time reinventing them, I think reading source code is a pretty useless exercise unless you also have a detailed record of how that source code came to exist in its present form. Until there is some revolutionary new tool for generating a human-understandable narrated history of large-scale design decisions from a source control history, your time will almost certainly be better spent reading textbooks that incrementally develop a piece of software over several chapters. Even that is cheating -- the authors know exactly where they want to end up and they won't include all the missteps they made when they first started writing similar programs. But it's still loads better than the alternative. Just as sitting in a law school library absorbing an encyclopedic knowledge of the law won't really train you to make arguments that will fly in front of a judge, reading a code base as a dead, unchanging document won't teach you what it is to live in that code.
It is possible to make misteps gracefully - that is, even when making mistakes you can mitigate the problems by writing clean code and trying to make it fit nice patterns you've already seen.
Sure, you can't copy directly, but you can get a feel for what is nice and isn't by looking at lots of code that has reached it's goal, and thus get an intuition for code smells.
Further, seeing how principles and patterns are used elsewhere helps take abstract concepts and turn them into real examples - for me anyway this is vital for my learning of anything.
