lme() vs aov(y ~ A*B + Error(aa %in% A + bb %in% B)) [repost]

I've posted the following to R-help on May 15.
It has reproducible R code for real data -- and a real
(academic, i.e unpaid) consultion background.

I'd be glad for some insight here, mainly not for myself.
In the mean time, we've learned that it is to be expected for
anova(*, "marginal") to be contrast dependent, but still are
glad for advice if you have experience.

Thank you in advance,
Martin