Are you all by any chance comparing scenarios where the market prices in the pre-buy-both and post-buy-both are at 50% 50%? For example, are you using [5, 5] shares first, and buy 45 of each for [50, 50] shares, and looking then at the effect of buying one of the first share [+1, +0] on the market prices?
If you are, I would try it again with starting shares of [4, 6] and increasing to [40, 60].
Buying more shares of all states does have the same practical effect as increasing b, but the market scoring rules (logarithmic, spherical, quadratic) change the expression of that practical effect. The LMSR pumps in some extra 'practical liquidity' by draining liquidity from the center (".5", where it assumes individuals will be eager to trade) and dumping it into the edges.