So you have two samples, each of which consist of a number of scores as percentages? How big are the samples?

It is true that the distribution of scores within each sample cannot be completely normally distributed given that percentages are bounded at 0 and 100%. But with a reasonably large sample, the sampling distribution of the means may be close enough to normal as to make no difference. It's the sampling distribution of statistics that we usually need to worry about the most, not the distribution of data points themselves.