Massive multiple-input multiple-output (MIMO) relies on accurate channel estimation for precoding and receiving to achieve its claimed performance advantages. When serving vehicular users, the rapid channel aging effect greatly hinders its advantages, and a careful system design is required to ensure an efficient use of wireless resources. In this paper, we investigate this problem for the first time in a non-isotropic scattering scenario. The von Mises distribution is adopted for the angle of arrival (AoA), resulting in a tunable channel temporal correlation coefficient (TCC) model, which can adapt to different AoA spread conditions through the k parameter and incorporates the isotropic Jakes-Clarke model as a special case. The simulated results in a Manhattan grid-type multi-cell network clearly demonstrate the impact of channel aging on the uplink spectral efficiency (SE) performance and moreover, in order to maximize the area average SE, the size of the transmission block should be optimally selected according to some linear equations of k.