For large-scale three-dimensional electromagnetic problems with fine structures in two spatial directions, the Courant-Friedrichs-Lewy (CFL) condition imposes a severe restriction on the finite-difference time-domain (FDTD) method. Its time step is limited by the smallest grid size, resulting in low computational efficiency. The weakly conditionally stable FDTD (WCS-FDTD) method relaxes this restriction through implicit field updates. However, large tridiagonal systems must be solved repeatedly at each time step. These solutions impose substantial computational and memory demands on a single computing node. To overcome these limitations, an MPI-based parallel three-dimensional WCS-FDTD method is proposed for distributed-memory platforms. The field components along the two finely discretized directions are updated implicitly. Therefore, the time step is determined by the grid size in the coarsely discretized direction. The computational domain is decomposed one-dimensionally along this coarse-grid direction. As a result, each MPI process retains complete grid lines in both implicit directions. It independently solves the local tridiagonal systems using the Thomas algorithm. No interprocess communication is required during the implicit solutions. After the corresponding field updates, neighboring processes exchange only the required boundary field components through MPI Sendrecv. This strategy combines the enlarged time step of WCS-FDTD with process-level parallelism. It also distributes the computational workload and field data across multiple nodes. A metallic metagrating is used to evaluate the accuracy, computational efficiency, and parallel scalability of the proposed method. The calculated near-field probe responses agree closely with those obtained using the conventional FDTD and serial WCS-FDTD methods. Close agreement is also observed for the far-field radar cross sections. For this example, WCS-FDTD increases the time step from 0.70 ps to 6.67 ps. It reduces the serial runtime by a factor of 3.8 relative to conventional FDTD. When the same 12 MPI processes are used, parallel WCS-FDTD is 4.2 times faster than parallel FDTD. It is also 11.5 times faster than serial WCS-FDTD. With 32 processes, the parallel WCS-FDTD method achieves a speedup of 30.5 and a parallel efficiency of 95.3%. Multi-node tests are also performed using 16 MPI processes per node. Increasing the number of nodes from one to three reduces the runtime from 223.2 s to 89.4 s. This corresponds to a relative speedup of 2.5 and a cross-node parallel efficiency of 83.2%. These results demonstrate that the proposed method preserves the accuracy of WCS-FDTD and substantially improves computational efficiency. It also effectively exploits the aggregated computing and memory resources of multiple nodes. Therefore, the method is well suited to large-scale three-dimensional electromagnetic simulations involving fine structures in two spatial directions.