Coherence messages in token protocol are always broadcast, which creates heavy network traffic and limits its scalability. An efficient predictor technique was used to eliminate the useless message in token protocol, which employs a predictor to predict the data’s sharers and only sends coherence messages to the sharers to avoid broadcasting. Three predictor policies depending on the write miss and read miss ratio of application were proposed. The results show that: the owner predictor with 512 entries can reduce interconnect traffic by 3.8% on average, the sharer predictor and the hybrid reduce interconnect traffic by the average of 11% and 7% respectively. The method proposed efficiently reduces the traffic and improves the scalability of token protocol. 