The Radio Access Network (RAN) of Fifth Generation (5G) communication systems critically influences the overall performance and cost of a 5G system. 5G RANs allow for a gamut of implementations ranging from closed (carrier-grade) implementations by traditional vendors that are expensive and configured by the vendor to RANs that are compliant with the specifications of the O-RAN ALLIANCE, i.e., to so-called O-RANs. O-RANs are available in closed-source and open-source implementations; O-RANs can be configured by the operator and are cheaper than the closed RANs. Understanding the packet-level performance characteristics of these different 5G RAN implementations is a prerequisite for informed network planning and deployment, particularly for Non-Public Networks (NPNs) for specific use cases. Existing measurement studies have mainly considered a single User Equipment (UE); however, practical networks typically operate with multiple UEs. We conduct packet-level measurements for concurrent over-the-air transmissions to (downstream) and from (upstream) multiple UEs for a wide range of packet transmission rates. We consider current representative implementations of three RAN architectural options: closed RAN (based on Nokia system), closed-source O-RAN (based on Airspan system), and open-source O-RAN (based on srsRAN Project system). We find that the one-way packet delays and the packet loss probabilities can vary sharply among the multiple concurrently operating UEs (which are identical in model and configuration) in the closed RAN and open-source O-RAN; whereas, the packet-level performance tends to be more consistent among the UEs in the closed-source O-RAN. Our measurement results provide a baseline for future research on refining RAN operations for multiple concurrent packet flows.