In this paper, a cellular-connected unmanned aerial vehicles (UAV) mobile edge computing system is studied in which the multiple UAVs are served by terrestrial base station (TBS) for computation task offloading. Our goal is to minimize the total UAVs energy consumption, including propulsion energy, computation energy and communication energy, while ensuring that the total number of bits of UAVs are completely computed. For tackling the large amount of bits for computation, we propose a resource partitioning strategy where one portion of tasks is migrated to TBS for computation and the other portion of tasks is locally computed at UAV. For deeply comprehending the impacts of access manners on the system performance, we consider four access schemes in the uplink transmission, i.e., time division multiple access (TDMA), orthogonal frequency division multiple access (OFDMA), One-by-One access and non-orthogonal multiple access (NOMA). The problem of jointly optimizing bit allocation, power allocation, resource partitioning as well as UAV trajectory under TBS's energy budget is formulated and tackled by means of successive convex approximation (SCA) technique. The numerical results show that the proposed schemes save much energy compared with benchmark schemes.