Background: When guiding a remote collaborator in a virtual environment, people often take an addressee-perspective, which may have a high cognitive cost. In order to improve collaborative virtual environments, a better understanding of how operators share spatial information is needed. Aims: This work aimed to study the cognitive workload linked to spatial statements production in situations in which the relative positions of speaker, addressee and target were varied. Method: Twenty-two participants were asked to give - in one go - instructions to a virtual collaborator on how to find a target in a 3D environment. The scene showed an avatar in the center of eight tables. Sixty-four configurations of avatar orientation (eight possibilities) and target location (on the eight tables) were tested. We measured the delay in starting the instruction once the target appeared, the instruction duration, and the subjective evaluation of mental demand. Each instruction was classified according to the spatial reference frame used. Results: The delay was influenced by the processing of spatial information in ego-centered and addressee-centered reference frames. All subsequent measures were determined by mental transformations in addressee-centered coordinates. One condition in particular, when the target was situated diagonally behind the addressee, gave rise to a higher mental demand for the speaker, which points to the investment made by the speaker in achieving the least collaborative effort. Conclusions: Further work should seek to develop efficient tools to facilitate spatial communication in situations that induce the most mental workload.