The problem of similarity searching is nowadays attracting a lot of attention, because upcoming applications process complex data and the traditional exact match searching is not sufficient. There are efficient solutions, but they are tailored for the needs of specific data domains. General solutions, based on the metric space abstraction, are extensible, but they are designed to operate on a single computer only. Therefore, their scalability is limited and they cannot adapt to different performance requirements. In this paper, we propose a distributed access structure which is fully dynamic and exploits a Grid infrastructure. We study properties of this structure in numerous experiments. Besides, the performance tuning is analyzed with respect to user-specific requirements which include the maximum response time and the number of queries executed concurrently.