To improve traffic throughput, Cooperative Adaptive Cruise Control (CACC) has been proposed as a solution. The usage of Vehicle-to-Vehicle (V2V) communication enables short following distances, thereby increasing road capacity and fuel reduction (especially for trucks). Control designs for CACC use the wirelessly communicated intended acceleration of a preceding vehicle as a feedforward action in a following vehicle. This feedforward action may determine approximately 80% of the total control action. In case of a communication failure, this feedforward is no longer available, and a larger time gap is needed to ensure high performance and robustness in terms of stability and safety. However, such a larger time gap is not instantly realizable. Therefore, a CACC design is needed which is robust against intermittent communication failures. This paper proposes to share model-based predictions of the intended acceleration via V2V communication, which are stored in a buffer of the following vehicle. This buffer is used in case a packet dropout occurs. Further, since the communication frequency is lower than the frequency of the control-platform, this buffer is also used to virtually upgrade the communication frequency. The design has been tested in experimental vehicles and shows an increased control performance, also in periods of packet dropouts.