In this article, we introduce and evaluate several new algorithmic improvements to Ray Hierarchies. The improved algorithm is used to produce ray-traced shadows for omni-directional lights in fully dynamic environments. The new algorithmic improvements increase culling rate of the traversal algorithm, in some cases dramatically cutting the number of visited nodes. To evaluate performance, we present a GPU implementation using CUDA, which uses a hybrid breadthdepth algorithm to perform the traversal using bounded memory. The results show that the use of improved cone hierarchies is able to produce high quality shadows quickly, for some scenes in real-time.