Network Function Virtualization (NFV) is envisioned to provide the foundation for a dynamic, high-performance multi-layer network architecture. With NFV, running network functions on general-purpose hardware is supposed to lower network latency due to reduced propagation delay. The tradeoff, however, is in impaired reliability and increased processing latency, which could negatively affect Quality of Service (QoS) and Quality of Experience (QoE). We investigate the tradeoff via a comprehensive measurement study exploiting a high-precision measurement method on a carrier-grade Cisco Catalyst 9500 Series switch, which provides enhanced computing capability via lightweight containerization. Baseline results on packet forwarding latency confirm that software-based forwarding is inferior to the hardware pendant. Subsequently, we deploy a more realistic application with an example for network coding and introduce a new metric, re-coding opportunities. Our study identifies a superior performance of generic hardware, but also shows advantages of in-network computing for latency-tolerable scenarios.