I'm interested to hear how the LMSR performs in practice - I've read a report that says although it has bounded loss by design, in practice it nearly always looses.
Considering that it only profits if the outcome is contrary to what the market predicts at the very end, I would expect this to be true. The point of an LMSR is mostly that the initial liquidity provider is not doing it to make profit but rather altruistically for the good of getting accurate predictions on the question being asked.
However, if the market maker charges trading fees that go to the initial liquidity provider, then profit becomes very possible. In fact, since the loss is bounded, after some amount of trading occurs, the initial liquidity provider will break even and anything after that is pure profit. So if the initial liquidity provider is not creating the prediction market altruistically, they need to bet on whether their particular PM will generate enough trading volume to make back their investment.
Also, a liquidity sensitive LMSR is an improvement on regular LMSR that allows some fraction of the trading fees to instead go back into the liquidity pool. This means that with more trading the liquidity of the market automatically increases (slippage decreases).