If your n per group is 4, then you can run paired post hoc tests, but I wouldn't use Dunnett's as it is a special case of a t-test, which implies at least interval strength data. If your data are only ordinal strength, then a test such as the Mann-Whitney U would be an option (comparing two batches at a time), or the Kruskal-Wallis anova by ranks method (for comparing two or more batches at a time) could be used as well. There are numerous ways to adjust the aggregate risk of type I error due to running multiple tests.
However, if the total N = 4, you'll be hard pressed to show differences, even with only two groups, no matter what statistical test you try.