I agree with Joseph Nolan's reply.
Additionally, it seems an odd choice of a variable upon which to stratify your data. What effect does "number of records" tend to have on your response variable of interest? While I am obviously uninformed about the nature of the variables you are working with, on its face it seems that there could be better choices upon which to stratify. When selecting a variable upon which to stratify, you need, at minimum, a theory as to why each of these strata may tend to have very different variances and response variable findings.
------------------------------
Gretchen Donahue
------------------------------