This saves a function call by using subclassing dict instead of using a
real dict. When a cache hit occurs, there is no overhead beyond a
standard dict lookup, which in most implementations is very fast. Cache
miss is similar performance to previous.
Also added a unittest for this functionality.