Making GIS and R play nicely - some tips?

Hi R Spatial People:

I am an experienced R user and an experienced GIS user - however, that GIS
is Arc. In the past when I've worked with spatial data in R, I've done so by
exporting the GIS data (always grids) from Arc as ASCII files. Then I read
them into R and go through some gymnastics to dump the no data values
(-9999) while I do my analysis (say a regression tree) and then write the
ASCII back out and suck it into Arc for display. This lose coupling seems
inelegant.

I'm starting on a new project where I have a time series of about 250 grids
that are roughly 1000 by 1000. There is a response variable and up to six or
so predictors. So, it's enough data that I feel it's time to learn how to do
it right. Question is, what's right?

Is the state of the art approach to use GRASS a la Bivand and Neteler's
paper (http://agec144.agecon.uiuc.edu/csiss/Rgeo/)? Is it better to dump the
grids to an imagine file (or bil) and read them with rgdal? I work mostly in
Windows (b/c of ESRI - damn them!) but switch gears onto Linux when the task
requires it. I have never used GRASS. Show me the way!

Thanks in advance, Andy