An increasing demand for geographic data compels data providers to handle an enormous amount of range queries at their data tier. In answer to this, frequently used data can be cached in a distributed main memory store in which the load is balanced among multiple cache nodes. To make appropriate load-balancing decisions, several key-indicators such as expected and actual workload as well as data skew can be used. In this work, we make use of an abstract mathematical model to consolidate these indicators. Moreover, we propose a multi-level load-balancing algorithm which considers the different indicators in separate stages. Our evaluation shows that our multi-level approach significantly improves the resource utilization in comparison to existing technology.