Boston Naming Test

Dear Friends,

I am trying to define the differences of two means in Boston Naming Test (BNT). I want to conduct a multipl regression with education and age as covariates. However, The raw scores of BNT are very skewed due to the high ceiling effect. I have found heteroscedasticity in my results. Log did not work either. And I have no chance for z-scores. Any suggestions? Do you think bootstraping helps this problem?
Or put it in an another way, what is the general way for adjusting age-edu effects when we brake the distribution-homoscedasticity assumptions of regression?