Collaboration in virtual reality (VR) across heterogeneous devices poses challenges for effectively supporting manipulation, navigation, and communication through different interfaces. We explored these in the design and development of a collaborative VR system that allows interaction through a mobile touchscreen device (Handheld User) and a head-mounted display (HMD User). In a qualitative evaluation, we further analyzed how these interfaces affect social roles and user interactions during collaboration. Our observations reveal that Handheld Users achieved presence in the virtual environment, despite the non-immersive interface, and assumed similar spatial positions in a conversational scenario as they would in the real world. In a collaborative building task, we found leadership roles not tied to immersion, but potentially influenced by users’ eye-level. Further, Handheld Users exhibited stronger movement patterns than HMD Users. Based on such behavioral patterns, we contribute a classification framework for Handheld Users that facilitates future analysis of interactions in shared virtual environments through handheld devices. Finally, we offer several design considerations for collaborative cross-device VR, which are based on our observations and exemplified in our presented system.