Interestingly, seems like experts and masters were not very motivated solving all problems, especially easy ones.

Problem difficulties were calculated as described here or here. Basically, it means that if problem difficulty (rating) equals to yours then there is 0.5 probability that you will be able to solve it during contest. Or, alternatively, rank of user with such rating would be equal to number of participants who solved it. To calculate problem rating you could use Maximum Likelihood Estimation method or just binary search.