When developing systems of systems, requirements tend to be redundant especially when running large numbers of projects, with many requirements per project, and diverse sources of requirements. It is therefore necessary to consolidate requirements by identifying the ones that are equivalent in order to avoid redundant work. The aim of this paper is to evaluate requirement similarity measurement to support analysts when linking equivalent requirements. The evaluation is conducted based on the requirements management process of an Italian company in the defense and aerospace domain. Our empirical investigation combines a controlled experiment with graduate students and an industrial case study. Results clearly show that one cannot expect any significant advantage in general. The level of support provided by similarity measures significantly depends on their level of credibility, that is the extent to which similarity measurement reliably indicates the equivalence of requirements. On average, given the credibility distribution observed in our industrial case study, showing similarity measurement to analysts is expected to: 1) improve by 20% the number of equivalence links identified per minute and 2) decrease by 40% the number of incorrect links. Finally, we investigate whether there is an effective way to combine human judgment and similarity measurement to effectively determine equivalence links. Based on machine learning, our approach yielded positive results both in terms of the correctness of the links and the speed at which they are established. Moreover, this hybrid solution is effective even when the credibility of similarity measurement is half the average we observed in our industrial case study. In conclusion, our results confirm and complement past empirical studies on the practical benefit, in terms of both quality and speed, of adopting requirement similarity measurement for linking equivalent requirements.