The last decade has seen location-based services as the rising star on the firmament of pervasive computing. Infrastructure-based middleware platforms provide a scalable way to support those applications. At the mobile side, event-driven control flow dominates the software design. Location events like "User A is NEAR user B" and their logical combinations must be processed effectively. We suggest a scalable event filtering strategy by separating the evaluation of basic elements from the computation of the remaining logical expression and by dynamically distributing the logic-computing unit throughout the platform infrastructure. Distribution is made according to the user's positions. In the best case, this leads to fully distributed processing. In the worst case, the scheme falls back to a quasi-centralized design. Measured performance of a prototype implementation in Java was at 355 notifications per second. Compared to a centralized design, this is an improvement by roughly a factor of 70.