Most space surveillance and tracking systems are constituted by networks of ground stations of observing radars and optical telescopes. These systems are usually reliable, easily serviceable, and effective under ideal conditions, but suffer from strong bounds on scalability and coverage due to the heavy constraints on their geographic locations, potential cloud coverage and the perturbing effect of the atmosphere. A large constellation of small satellites carrying optical telescopes could complement these limitations, thanks to the lack of such constraints and the possibility of observing target objects at close range. As a result, it would be theoretically possible to track objects more accurately and for longer times, improving the accuracy of collision risk analysis and manoeuvre detection amid other tasks. However, due to the small fields of view of suitable onboard optical sensors, a random static arrangement of their lines of sight would be largely inefficient in reaching good performance levels, as the target would unpredictably enter and exit the observable portion of the sky. To solve this problem, we propose a cooperative intelligent tracking strategy for the constellation. Assuming known initial states for some targets, we use predictions on the future states to dynamically control the attitudes of the constellation satellites to maximise the length of the tracking window while minimising energy expenditures. We evaluate the performance of the strategy using quality figures such as the number of effectively trackable targets and the mean square error of the estimation error during tracking. We repeat the analysis for various constellation geometries and multiple target orbits to investigate the general applicability of such a strategy. In conclusion, we check for robustness by analysing performance drops under the loss of operating nodes. The results thus obtained will inform on the usefulness of space-to-space SST constellations and the general design of strategies for the dynamic scheduling of operations of distributed systems observing multiple targets.