Gas–liquid density ratio (DR) is a key dimensionless number in sloshing assessment methodologies of membrane containment systems for LNG tanks of floating structures. Earlier studies on the effect of DR were mainly statistical and effects of DR were usually mixed with those of gas compressibility and ullage gas pressure but attributed only to DR. In an attempt to separately study such effects, Karimi et al. (2015) [11] studied the effects of DR far from impact zones (global effects of gas–liquid density ratio) which proved to be small in the studied range of DR (0.0002 to 0.0060). The effects of DR near impact zones and before detection of any compressibility effects are referred to as local effects and correspond to modifications of wave shape before impact. They were treated in Karimi et al. (2016). This paper studies the influence of ullage gas at the same scale as well as scaling of sloshing loads at different scales. The test setups were similar to those presented in Karimi et al. (2015) [11] and Karimi et al. (2016) and consisted of three 2D model tanks as transverse slices of tank 2 (out of 4) of a membrane LNG carrier with total capacity of 152000 m³ at scales 1:10, 1:20 and 1:40. All model tests were performed at a fill level corresponding to 20% of the tank heights. Water as liquid and different ullage gases of helium (He), air, two mixtures of sulfur hexafluoride (SF₆) and nitrogen (N₂), and pure SF₆, all at atmospheric pressure with a range of DRs from 0.0002 to 0.0060 were used. Synchronized High-speed video cameras (@4000 fps) and arrays of piezo-electric PCB pressure sensors (@40 kHz) monitored and measured impacts on the tank walls. The study was mainly based on the definition of Impact ID based on impact coincidence. The results are presented at 4 main stages. First, in the same way that sloshing loads measured in irregular model tests are treated in the current methodologies, the measured pressure peaks are studied as statistical samples. Next by the notion of impact ID, the effect of change of ullage gas at the same scale is verified. Thirdly with the same notion of impact ID, impacts are tracked down through three scales to verify scaling. At last dominant impact IDs are introduced. It is shown that the most severe impacts are generated by only a few dominant IDs.