If you want to track differences between different studies, you should not compare the posterior plots directly. You should keep track of it via something like
dif_mu = pm.Deterministic("dif_mu", mu - mu.T)
Here dif_mu[i,j] will give you the difference between mean of study i and j. Don’t mind the warnings about scalar divide, they happen because mu[i,i] - mu[i,i] is always zero.