Optimum cell size does not depend on speed of mobile. It depends mainly on path loss parameters. The boundary of cell limited by inverse fourth power law (you can check the derivation based on two ray model)

The maximum radius of the cell is limited by power loss. But the number of users are limited by the bandwidth. Since bandwidth is a constraint, frequency reuse is recommended. The smaller the cell radius, the better frequency reuse and hence more subscribers. Therefore, by reducing the cell size, we are able to connect more number subscribers within the same spectrum. 

But what should be the minimum cell size ?

Obviously, the limitation should come from the handover, which in turn depends on handset velocity. 

Minimum cell size can be any value, depending on type of application, service, transmit power, number of user, user QoS and bandwidth. The size of cell varies from Macrocell (5 km) to Attocell (1 m) or may be less. The minimum cell size is limited in different network by different parameter. In case of cellular, inter carrier interference (frequency reuse), speed of mobile (eventually number of handover) are the two main parameters, whereas data rate and transmit power in WiFi, and SE (8 bps/Hz), interference in femtocell etc. 

Actually in radio planning of any area, the network operator decides the number of cells either for capacity-limited design (large cells - small number) or coverage-limited design (small cells - large number). For handover in a heterogeneous network, the mobility level of the user is considered such that for example the high speed user (>100 Km/h) is suspended from handover to small cells in order to avoid Ping-Pong and handover failure states. For medium speed users (>60 km/h) the user is guided by the large cell (Macrocell) to initiate handover at certain distance from the small cell in order to save power spent on cell search.