I have been looking into this for a different use case.  I have not seen it in the situation you describe.  I tried using your script there driven with a csv of 50 products with random data. For me, in each try there were no duplicate products create...