Dear Wilhelm
There is an article, including a part about fitting linear mixed
models. There the problem with the degrees of freedom is
described.
You can have a look to the second link, too, discussing the
problem as well.
http://cran.r-project.org/doc/Rnews/Rnews_2005-1.pdf
I have a problem moving from multistratum aov analysis to lmer.
My dataset has observations of ampl at 4 levels of gapf and 2 levels of bl
on 6 subjects levels VP, with 2 replicates wg each, and is balanced.
Here is the summary of this set with aov: