Achieving high reliability in the presence of fading is particularly challenging under latency constraints, because the usual way of error mitigation by repetition becomes unfavorable. On the other hand, multi-connectivity does improve reliability without adding latency, but multiplies the required bandwidth per link and does not scale to a large number of users. There is hence a need for frequency diversity in a spectrum-efficient way. In this work, we investigate multi-user resource allocation schemes both without and with knowledge of each user's channel state. We evaluate the allocation-dependent reliability in terms of outage rate and outage duration based on simulations of automated guided vehicles in an industrial environment. To increase validity and ensure real-world correlation between vehicles, we draw channel states from high-resolution channel measurements at a factory floor. For channel-aware allocation, we propose a near-optimal low-complexity algorithm using different quality functions based on channel state preference lists. Since accurate channel information per user and resource incurs signaling overhead, we also evaluate the algorithm's sensitivity to the number and bandwidth of resources as well as to outdated channel information. In conclusion, channel-aware allocation offers significant reliability improvements over static allocation and emerges as a key enabler to realize ultra-reliable low-latency communications on a larger scale.