Legged manipulators that combine legged robots with manipulators have emerged as a new solution to large-scale structure on-site manufacturing tasks. However, the productivity of legged manipulators is strongly affected by the positioning and repositioning process. This paper aims to improve the productivity of legged manipulators by reducing the required positioning and repositioning times. By treating a legged manipulator that attaches all feet to the ground as a virtual parallel robot, the workspace of the on-board manipulator can be enlarged while the end-effector maintains the motion precision. By choosing working positions that enable a legged manipulator to cover more processing regions with the enlarged workspace at a time, the required positioning and repositioning times can be reduced. The proposed method is implemented through a variant of a genetic algorithm to find the minimum set of working positions that enable a hexapod manipulator to accomplish welding tasks on a large-scale structure. Simulation results show that the required working positions of the hexapod manipulator are reduced significantly and thus higher productivity can be obtained. The flowchart to implement the proposed method is expected to improve the productivity of other kinds of legged manipulators that perform on-site manufacturing tasks.