I dont see how you could correct that your data is not balanced, and that balanced data is an assumption for AOV
I can only think that you might gather more data under a balanced experimental design so that you have balanced data. I'll be interested myself to see if Stats experts see this thread and offer other alternatives.