You might look towards using a modeling framework designed for large
clustered data sets--hierarchical linear modeling or HLM (Raudenbush is
an important figure). Although these methods have been developed in
educational psychology statistics, they apply to nested data generally.
There is a software package called HLM that has a GUI and is easy to
set up and run. You can also look at literature on multilevel models.
You might do well to also consider using software that represents the
data using a sparse matrix such as in NLME or LME. See Bates' textbook
on this software package available in R. LME4 might be a little more
speedy. I don't have experience with SAS Proc Mixed, but that also may
not require a design matrix size as is required by Stata's ANOVA
routines for mixed models.
