For clarity, are you using a conversion of mrsos(0-10) = swvl1()(10/7)/1000 (kg m-2)? or mrsos = swvl1* 100? The former comes from Section 2.4 “Harmonisation of variables across catalogues” of your reference link, however it seems wrong to me and I believe it should be the latter.
mrsos uses 10cm depth, while swvl1 uses 7cm, so you either extend the swvl1 values for another 3cm to match the depth of mrsos or you reduce the mrsos values by 30% to match the swvl1 depth.
Also, mrsos is the mass of water in all phases states, while swvl1 seems unspecified and could be the liquid phase equivalent volumetric density. Perhaps the mass density is significantly off in locations where surface temperatures are cold and there is a significant amount of frozen water in the top 7cm, thus reducing the mass density of the water component of the soil? Are the mrsos values always lower than expected?
Edit: My supposition is invalid. swvl1 explicitly includes both liquid and solid phases, as stated in this post. Perhaps a strong vertical gradient in the moisture can shift the 0-7 and 0-10cm averages by a lot? They’re also products from entirely different models and grids, so there is bound to be some difference (50% is not yet an order of magnitude)