How would you adjust for the ceiling and floor effect when doing repeated measure t test?

I'm doing an experiment on memory and the range of score participants could get is from 0 to 9 (discrete count data). We are comparing their memory before and after treatment. However, about 50% of the participants get 9 out of 9 (ie perfect memory) before treatment. This ceiling effect means that there is no way for these participants to improve after treatment. Is there a way for me to adjust for that in my statistics? Would non parametrics help?How would you adjust for the ceiling and floor effect when doing repeated measure t test?I guess I would try to use the percentage of the people that achieve perfect scores as the metric. This metric does not have the small number problem that dealing with individual scores does. One could also look at percentage of people that score above a certain score as the metric.

One can also see what has happened to the scores of people who originally had a particular score before treatment and and after treatment. Obviously the people with perfect scores have no where to go but down. They are probably of interest in determining what reliability to place on the particular increases which occurred. It also is possible that a particular treatment might diminish the ability to remember.