Smart-city applications increasingly rely on distributed sensor infrastructures that continuously collect and process potentially sensitive data. Homomorphic encryption provides a promising approach for protecting such data by enabling computations directly on encrypted values, but its computational overhead remains a concern for resource-constrained edge devices. This paper investigates the practical feasibility of additive homomorphic encryption for smart-city sensor data processing using a baseline implementation of the Paillier cryptosystem without hardware-specific optimizations. A distributed prototype was implemented on Raspberry Pi devices, in which sensor readings are encrypted locally, combined in the encrypted domain, and decrypted only at the sink node. The prototype was first evaluated using a complete end-to-end aggregation workflow on Raspberry Pi 3 and Raspberry Pi Zero 2 W devices. Subsequently, the cryptographic operations were evaluated in detail for key sizes of 1024, 2048, and 4096 bits on Raspberry Pi 4 devices using real-world smart-city sensor datasets. The results demonstrate a strong dependency of execution time on the selected key size, while differences between the evaluated sensor datasets have only a minor influence on cryptographic performance. A comparison with RSA quantifies the additional computational cost introduced by Paillier encryption and its additive homomorphic functionality. Finally, practical sensor processing rates are derived from the measured execution times to relate the cryptographic overhead to realistic sensing frequencies. The results show that, despite its considerable computational cost compared with conventional public-key encryption, Paillier-based aggregation can provide a practical privacy-preserving processing approach for smart-city sensing scenarios with moderate sampling frequencies.