Performance of e-business applications strongly depends on internal implementations of persistent database structures and on effective algorithms processing these structures. Commercial database systems, which are the basis of e-business applications typically implement logical database structures with one-size-fits-all persistent storage structure. This work investigates a new class of database systems where a conceptual and logical view of a database can be implemented in many different ways depending on the performance requirements of database applications. We consider the improvements to performance of e-business applications through automatic changes of the persistent database structures in the ways indicated by the performance statistics obtained from tracing of the applications. In particular, this work describes a language for formal specification of variable persistent database structures, it shows how to find the best implementation structures for a given set of e-business applications, and it shows how to automatically transform one persistent data structure into another. We also investigate a problem how to automatically adjust the implementations of applications to the new persistent database structures.