The point of this class is to lower the memory footprint and speed up comparisons (pointer equality) in cases where an application stores multiple copies of the same object in various maps.
Since uniquing is performed by storing an immutable copy of the original object in a map until there are no further references to that object, it's pointless to use this uniquing unless the application would be storing at least two copies of the object.
Also, since this is thread-safe there is a lock management overhead wherever a uniqued object is released, so performance gains are to be expected only if the uniqued object has a relatively long lifetime and is tested for equality with other instances frequently.
In short, use with care; while uniquing can have a big performance advantage for some programs, this is actually quite rare.

The internal implementation of the uniquing works by taking immutable copies of the objects to be uniqued, storing those copies in a hash table, and swizzling their class pointers to a sub-class which will automatically remove the instance from the hash table before it is deallocated.
Access to the hash table is protected by locks so that uniqued objects may be used freely in multiple threads.
The name of the subclass used is the name of the original class with 'GSUniqued' added as a prefix.