When co-occurring with elevated levels of ambient relative humidity (RH), hot extremes are more perceivable and consequently more health-damaging. Quantifying changes in humid-heat extremes has therefore gained considerable scientific and societal attention, but a fundamental yet critical aspect to the estimation-data reliability-has been largely downplayed in previous analysis. By comparing similar to 10 observational and reanalysis datasets to fully-homogenized observations across China, we report ubiquitous inhomogeneity in RH series in these popularly-used datasets [including HadISD(H) and ERA5], which accordingly produce unrealistically strong drying trends 2-3 times the homogenized dataset-based estimate during 1979-2013 in warm-moist southeast China. Locally, an inhomogeneity-caused exaggeration of drying by a magnitude of 1% decade(-1) translates into a significant underestimation of increasing rates for frequency and intensity of humid-heat extremes by more than 1.2 days decade(-1) and .07% decade(-1) respectively. From a regional perspective, these inhomogeneous records have underestimated the frequency increase of extremes by up to 2 days decade(-1) and their intensification by up to .4 degrees C decade(-1) in southeast China. Extremes identified via homogenized and non-homogenized datasets also differ in the bivariate joint distribution structure, with former cases featuring similarly hot temperatures yet discernably lower humidity.