Ultra-Reliable and Low-Latency Communications (URLLC) has attracted considerable attention because of its potential applications in factory automation, automated driving, and telesurgery anticipated for the era of the sixth-Generation (6G) networks. In URLLC with random channel gains and a hard delay constraint, the scheduling of backlogged queues and finite blocklength coding in the physical layer make it rather challenging to specify its performance limit. In this paper, we focus our attention on the asymptotic cross-layer analysis of URLLC when the Signal-to-Noise Ratio (SNR) is sufficiently high. More specifically, we find that a fundamental tradeoff exists between the latency and error probability in the high SNR regime, which is characterized by a gain conservation equation. The main result of this work reveals that the sum of our defined real-time gain and reliability gain is equal to one under the optimal scheduling policy. Numerical simulations are also exploited to validate that the derived gain conservation equation holds even with bounded random arrival.