Hierarchical clustering Ward's method and normality assumption

[#1]

Hi, 

I have left censored data (values below detection are assigned zero) that I would like analyze by hierarchical clustering in JMP using Ward's minimum distance method.   I had assume that because Ward's method is based on "the distance between two clusters is the ANOVA sum of squares between the two clusters summed over all the variables." that the assumption of normality in the distribution should be satisfied. 

There are lots of zeros. 

Do I need to be concerned about the assumption of normality for hierarchical clustering?  That is, if my input data are means for the clustering, should those means be calculated so they are not biased because of the distribution?  I do the clustering on the means (or transformed means, or medians), which reduces some of the zeros but about a 1/3 of responses still have 40-75% zeros for the means of some sites, which I don't think this is a problem as that is the data structure.  

Thanks, BT

[#2]

There is no assumption about a normally distributed response. The statistical details about hierarchical clustering shows that the Euclidean distance is the measure that is based on the squared difference.