In this paper, we consider a large scale sensor network comprising multiple, say K, base stations and a large number of wireless sensors. Such an infrastructure is expected to be more energy efficient and scale well with the size of the sensor nodes. To support a large number of queries, we examine the problem of allocating queries across the base stations to minimize the total data communication cost among the sensors. In particular, we examine similarity-aware techniques that exploit the similarities among queries when allocating queries, so that queries that require data from a common set of sensor nodes are allocated to the same base stations. We first approximate the problem of allocating queries to K base stations as a max-K-cut problem, and adapts an existing solution to our context. However, the scheme only works in a static context, where all queries are known in advance. In order to operate in a dynamic environment with frequent query arrivals and termination, we further propose a novel similarity-aware strategy that allocates queries to base stations one at a time. We also propose several heuristics to order a batch of queries for incremental allocation. We conducted experiments to evaluate our proposed schemes, and our results show that our similarity-aware query allocation schemes can effectively exploit the sharing among queries to greatly reduce the communication cost.