In the paper, bike sharing systems with stations having a finite capacity are studied as large stochastic networks. The inhomogeneity is modeled by clusters. We use a mean field limit to derive the limiting stationary distribution of the number of bikes at the stations. This approach is an alternative to analytical methods. It can be used even if a closed form expression for the stationary distribution is out of reach as illustrated on a variant. Both models are compared. A practical conclusion is that avoiding empty or full stations does not improve overall performance.