Random vibration theory (RVT) site‐response analysis computes site‐specific site amplification without the need for the specification of input time series (TS). A fundamental parameter in RVT analysis is the duration used when calculating the root mean square (rms) oscillator response (i.e., spectral acceleration) from the Fourier amplitude spectrum. Recent research suggested that RVT site‐response analysis should incorporate the effects of site response on the duration used in predicting the rms oscillator response for the soil surface motion. Wang and Rathje (2018) developed an empirical model to predict the change in the duration of the oscillator response due to dynamic site response. This article evaluates the performance of RVT site‐response analysis using the duration modification presented in the companion paper (Wang and Rathje, 2018). It is shown that for a wide range of shear‐wave velocity profiles and earthquake scenarios, the duration modification produces RVT amplification factors (AFs) that, on average, are within of the AFs predicted by TS analysis. Similar improvement is obtained for linear‐elastic and equivalent‐linear site‐response analyses. The cases where the duration modification does not fully improve the RVT site amplification predictions are associated with low‐frequency sites subjected to small magnitude earthquakes, where the site frequency is less than one‐half the corner frequency of the input motion.