Weighted Reference Couting
Weighted reference counting is more efficient than traditional reference counting because you don't have to update the reference counted object every time a reference is taken (potentially meaning updating the object on the heap and blowing the processor cache). Weighted reference counting only updates the referenced object when a reference is dropped or when weight is exhausted. Weight Reference Counting is also really good for concurrent systems because the number of synchronisations needed is much lower than normal reference counting.
Allocate a new reference counted object with a default weight (a tunable power of two - usually around
1 << 16 for a real implementation). An initial reference is allocated which points to the object and the weight value is copied into it. Every time the reference is cloned, the weight is split in two (shifting one bit to the right is the same as dividing by two) and one half allocated to the existing reference and one