In recently introduced "multi-cell access" schemes, cells (rather than users) compete for the spectral resource. In a previous paper [1], a framework for such a scheme was proposed, using only local channel information. In the framework each cell competes for access through a credit that is a function of the signal to noise ratio of its scheduled user. Access is then given to a cell with a probability dependent on the access function. In [1] an ad-hoc choice of function was formulated. In this work we investigate which access function actually optimizes the system capacity, utilizing a numerical optimization procedure. We obtain a surprsingly simple solution which also corroborates previous results. For a realistic path loss model we find that our multicell distributed access scheme gives a 19 % gain compared to having all cells on, and more than a 50 % increase in system capacity compared to keeping a traditional static spectral reuse scheme.