A few days ago Facebook announced on its Engineering page that they decided to release as open source some of the C++ components propeling the most heavily loaded parts of their software infrastructure. This set of components is called Folly and can be downloaded from Github.
Folly code is mainly optimized for performance, so you can even find there Facebook's own implementations of such core C++ classes like string, vector and hashmap. All code is well documented and covered by unit tests.
However, there are some limitations to using Folly you have to remember about. First, the only supported platform so far is 64-bit Linux, and there are no plans of porting it to other platforms, yet. Second, Folly requires gcc compiler to work in C++ 11 mode, which means that as for today you need to provide -std=c++0x compilation flag in your Makefiles and make sure that your code complies with the new C++ standard.
Do you need Facebook Folly? In my opinion in most cases you don't. Remember that Facebook hires the smartest engineers they can find, so if they optimize native C++ libraries it means that they probably have no choice. For an average programmer it would be much more sensible to profile his or her code to identify and remove bottlenecks like suboptimal algorithms or overused I/O (too many disk operations or frequent memory allocations) and treat Folly as the last resort.