As long as you are happy with doing an analysis with the outliers…
Me personally, I would perform the analysis once with the outliers included and then with outliers excluded to see the difference. I would also look into the outliers and try to find out, why they occured (what are the possible reasons) to see if the results have some limitations on generalisability. If there is a significant difference, I would interprete the results without the outlayers and try to explain what happened with the outliers.
Your proposals are OK, you just find a way for the outliers to get closer to the mean (the low outliers get the next lowest -1). More outliers? The same score, why not?
You can also use some of the so called robust statistics, which are supposed to deal with outliers automatically…