LoRaWAN is a key enabler for Internet of Things applications, providing long range communication at a low energy consumption. However, extensive use of downlink communication can reduce network capacity and increase uplink loss, as shown in several studies. The network server is responsible for scheduling downlink packets, while respecting strict timing constraints on the end devices and on the gateways. This paper presents the first evaluation of downlink scheduling implemented by two widely used LoRaWAN network servers: ChirpStack and The Things Stack. Using the ELoRa emulation tool, we inject real LoRaWAN traffic into these network servers and analyze their downlink scheduling algorithms based on defined performance metrics such as reliability and schedule efficiency. Our findings indicate that The Things Stack has a more accurate scheduler, but more conservative, as it delivers less downlink packets. ChirpStack, on the other hand, with its simpler algorithm, performs comparably and even equals The Things Stack as the number of end devices increases.