Is Kruskal Wallis really needed?

#1
Dear all,

I really do not see why we need Kruskal-Wallis test at all. In terms of the distribution, only platykurtic distribution may increase type 1 error. When the break the homogeneity of variance we could use Welsh or Borwn-Forsythe. We have a posthoc test suitable for all types of data. We can also use bootstrapping. So, do you really think that KW test is really needed?
 
#2
In terms of the distribution, only platykurtic distribution may increase type 1 error.
Which test are you thinking about here?

Do you have a source for the following statement? ("only platykurtic distribution may increase type 1 error")