The terrestrial carbon and water cycles are primarily driven by photosynthesis and transpiration that are both regulated by plant stomatal conductance. Such a physiological link not only determines ecosystem functioning, but also modulates flux exchanges between the biosphere and the atmosphere. The relationship between water losses via transpiration relative to carbon gains via photosynthesis can be estimated by plant water-use efficiency (WUE), a metric that involves multiple definitions across leaf to ecosystem scales. Despite its various functional forms, forest WUE trends inferred from different approaches ubiquitously increase in recent decades. Although several mechanisms have been proposed to explain rising forest WUE at seasonal time scales, none of them have examined the intra-seasonal variability of WUE and its impacts on long-term WUE trends. Here, we analyze the statistical distribution of sub-seasonal WUE at 33 eddy covariance sites with at least 10 years of measurements to investigate the effects of short-term WUE variability. Our random-forest variable importance analysis suggests that recent increases in site-specific WUE observations are strongly correlated with the corresponding trends inferred from its 95^th percentile. Further, our results demonstrate that seasonal mean WUE correlates well (r = 0.75 to 0.89) with the number of most active hours (i.e., cumulated hours when WUE exceeds a site-specific percentile), highlighting the importance of short-term favorable microclimatic conditions. Collectively, our findings suggest that a proper representation of seasonal cycles in WUE is needed to mechanistically explain recent increases in WUE observations and improve estimates in terrestrial carbon and water cycles.