Distributed scene graphs are important in virtual reality, both in collaborative virtual environments and in cluster rendering. In Myriad, individual scene graphs form a peer-to-peer network whose connections filter scene graph updates and create flexible relationships between scene graph nodes in the various peers. Modern scalable visualization systems often feature high intracluster throughput, but collaborative virtual environments (VEs) over a WAN share data at much lower rates, complicating the use of one scene graph system across the whole application. To avoid these difficulties, Myriad uses fine-grained sharing, whereby sharing properties of individual scene graph nodes can be dynamically changed from C++ and Python, and transient inconsistency, which relaxes resource requirements in collaborative VEs. A test application, WorldWideCrowd, implements these methods to demonstrate collaborative prototyping of a 300-avatar crowd animation viewed on two PC-cluster displays and edited on low-powered laptops, desktops, and even over a WAN.