We present extensive numerical studies on the determination of coherent wavelength-division multiplexing (WDM) crosstalk penalties for ultradense wavelength-division multiplexed (DWDM) systems, focusing on carrier-suppressed return-to-zero (CSRZ) as well as on 67% duty cycle differential phase-shift keying (67% RZ-DPSK) at a spectral efficiency of 0.8 b/s/Hz. Our analyses reveal large statistical variations in the predicted required optical signal-to-noise ratio (OSNR) when changing the WDM channels' interference conditions, in particular their relative optical phases and their relative time shifts. The strong impact of the exact WDM interference conditions can lead to simulation inaccuracies of many decibels when using standard OSNR simulations techniques. In measurements of DWDM system performance, the long averaging time of bit error ratio (BER) test sets can hide these burst-error generating penalty variations, and may, therefore, lead to wrong interpretations, especially for systems employing forward error correction (FEC). To overcome the DWDM simulation problem, we introduce and thoroughly assess a new simulation technique that allows us to efficiently and accurately capture the average required OSNR penalty for DWDM systems with negligible statistical error.