Dual-image reversible data hiding (RDH) has become an important research direction in the field of information security due to its better security, embedding rate and image visual quality. Bhardwaj et al. proposed a dual-image RDH algorithm on encrypted domain to further improve the security, but the embedding capacity of this algorithm is limited. In this paper, we propose a new embedding strategy by decomposing each pixel of an image into three units and encrypting them using the Paillier algorithm. According to the improved maximum distortion control coding table, the secret messages are embedded dynamically using the averaging method, i.e., the number of bits of secret information is varied for each embedding. The experimental results show that the embedding rate of our method is improved by 0.72 bpp compared with the method of Bhardwaj et al. and 0.25 bpp compared with the original maximum distortion control method. At the same time, the original image can be recovered losslessly, so the reversibility is guaranteed. It can meet the dual requirements of privacy protection and secret communication in high-capacity scenarios.