The increasing numbers of available data sources have led to increased data redundancy and hence novel challenges for federations. Typically, federation engines query all endpoints that provide relevant data for a given query. However, considering the overlap, a subset of these sources might already be sufficient to obtain a complete answer. Further, we deliberately might not wish to include all sources in the evaluation and make a decision based the reliability of a source. We therefore present ORAQL (an Overlap and Reliability Aware Query Processing Layer), an approach that exploits statistics capturing the overlap between sources to choose a subset of the available sources in the federation to compute a complete answer while minimizing redundant answers. Moreover, a user-provided reliability goal is taken into account. Hence, we propose an approach based on a majority vote over multiple sources to increase the reliability of the query result. For this work, we focus on TPF interfaces, since they are the least expressive interfaces and hence our approach can be adopted for more expressive interfaces, e.g. SPARQL endpoints. The presented methods to capture the overlap between sources of a federation have shown to generate useful overlap profiles with a maximum deviation of less than five percent. Even if the identification of redundant data is NP-hard we presented an approximation with a significant reduction in requested endpoints. Further, we have shown that ORAQL is granularly tunable towards reliability and can beat a state-of-the-art baseline system in terms of coverage and reliability.