John's last commit to vfs_bio.c restored part of the old behaviour but
not all. Even with 20 MB RAM I keep on seeing the disk going berseck
during stats. Less often than before but its still there.
Part of what I am doing tonite is integrating/testing Bruce's vnode
change, and making an equivalent change to the object management. Hope
that these changes will help. Also, I might change the threshold for
the amount of meta-data storage allowed in vfs_bio. I'm working on it!!!