One of the advantages of localism that is frequently cited is that the diversity of local approaches to a given problem and the relative ease with which new approaches can be adopted provide more opportunity for better solutions to emerge. It provides opportunity for experimentation and innovation, and is analogous to how the Cambrian Explosion resulted in the selection for better-adapted morphologies, or how a shotgun strategy (or throw stuff at the wall and see what sticks strategy) in business allows identification of niches and consumer preferences.
However, a problem that can occur with localism is that local approaches remain entrenched despite inferiority. Or put another way, once a superior approach has been identified in one locality, it is not implemented in other localities (whether such implementation is a push from a centralized authoritiy, or a pull from other localities copying).
So how can the benefit of experimentation be implemented within a centralized structure? Whilst generally working to the (hopefully optimal) centralised approach, local administrative structures would be able to apply for autonomy on a specific problem in order to test a defined alternative approach against a robust set of criteria. As a result of the central control, pushing out a new approach once an experiment has been concluded is relatively easy - the homogeneity of localities means that a single method of implementation is required for the new approach.