Overall performance of a distributed system depends considerably on the amount of load information used for an automatic distribution of load units. This may concern processing requirements in order to optimize utilization-based load distribution, as well as data requests to exploit actual data distribution for an affinity-based scheduling. We investigate the information contained in commercial workloads of on-line transaction-based database processing, and usable for a load distribution algorithm for both utilization- and affinity-based load distribution. This information must depend on the few identifiers of each load unit visible a priori for the load distribution algorithm. The identifiers establish a partition of the workload the load distribution is based upon. Assuming a database-sharing-like data distribution architecture, we propose quality measures for this finest and coarser workload partitions. They give information of system performance improvement if the load distribution is done according to these partitions, primarily in terms of overall data references and data references to remote nodes of the system.