Windows are points of the lowest resistance in the total sound and heat insulation of buildings. To enhance both thermal and acoustical qualities of windows with insulating glass units (IGU), the gaps between the panes can be filled with a heavier-than-air gas alone (for example, argon, krypton, xenon, and sulfur hexafluoride) or mixed with air. Here, the sound transmission loss (STL) may be increased in the low-frequency range dominated by acoustic resonances where the panes and gaps play the role of masses and springs. This frequency region (typically 100-500 Hz) is important for reducing traffic noise by windows. However, the expected acoustical improvement is often not measured using the standard two-reverberation room techniques. In this paper, such a discrepancy is theoretically and experimentally studied for physical interpretation and design optimization. Clear closed-form expressions have been derived and verified for the direct (via the panes and gaps) and flanking (via the panes and perimeter structural links) sound transmission through single-space and double-space insulating glass units, and improvement proposals have been developed. The results can be interesting for the scientists and engineers engaged in acoustics and construction.