We focus on the problem of constructing collaborative virtual environments (CVEs) that scale to large numbers of simultaneous participants and yet which still afford rich and varied possibilities for communication. This is achieved by extending our previously defined spatial model of interaction for CVEs to include third-party objects that provide support for contextual factors in awareness calculations and that enhance scaleability. Third parties can have two effects on awareness: attenuation or amplification of existing awareness relationships, and the introduction of new aggregate awareness relationships. We propose a range of applications for third-party objects including world structuring regions, aggregate views, dynamic crowds of participants, common foci, representational and group services, and dynamic load management. We also discuss how the third-party concept relates to other approaches to structuring virtual environments such as tiles, zones, and locales. We then present an implementation, the MASSIVE-2 system, focusing on its network architecture which is based on a dynamic and selfconfiguring hierarchy of multicast groups. Finally, we describe four demonstration applications that have been developed in MASSIVE-2: an environment for staging a public poetry performance that includes semiprivate zones for social interaction; the Panoptican Plaza, which demonstrates a variety of differently bounded regions; a collaborative 3-D Web browser that groups pages into server regions; and the Arena, a performance space that supports both static and mobile crowd aggregations of groups of participants.