Data grids are current solutions to the needs of large scale systems and provide a set of different geographically distributed resources. Their goal is to offer an important capacity of parallel calculation, ensure a data effective and rapid access, improve the availability, and tolerate the breakdowns. In such systems, however, these advantages are possible only by using the replication technique. The use of this technique raises the problem of maintaining consistency of replicas of the same data set. In order to guarantee replica set reliability, it is necessary to have high coherence. This fact, however, penalizes performance. In this paper, the authors propose studying balancing influence on replica quality. For this reason, a service of hybrid consistency management is developed, which combines the pessimistic and optimistic approaches and is extended by a load balancing service to improve service quality. This service is articulated on a hierarchical model with two levels.