Can you suggest a single use case for putting /usr on a separate partition that wouldn't work far better without? Sharing /usr across the network makes no sense, since its contents need to come from the same packages and versions as /.
My recollections may not be totally accurate, but I'll try.
In 1997, I was working at TCSI. My SPARC 5 workstation had a small local disk, which had crucial things like vi on it, which were fast to access. It might have been 500 megabytes or less. /usr was NFS-mounted over an unswitched 10-megabit Ethernet to a gigantic 35-gigabyte NFS server, from which I had access to a huge variety of software, including something like 8 versions of the CenterLine C++ compiler, much more than would fit on my local machine, but which was much slower to access. My home directory was also NFS-mounted.
Installing the huge variety of software on everybody's workstation would have meant that we had to have dozens of copies of it, using up precious disk space, and furthermore that installing or updating software would need to be done on every single workstation, instead of just on the NFS server. rdist existed but wasn't widely used; rsync had just been invented.
Also, "packages and versions"? pkg_add isn't what you think of as a package management system. The FHS was adapted from Unix workstation practice established before Debian invented package management as we know it today, and at a time that Slackware still didn't have package management. You could install your packages wherever you wanted, much as people do today with easy_install or gems.
A homogeneous environment that doesn't have formal package management is the only use case I can envision.
It was a good idea back in the SunOS days (and it's probably still workable to a limited extent for the BSDs, and perhaps folks like Gentoo) when local storage wasn't as plentiful as it is today, but I think we've safely moved on at this point.
On modern Solaris systems, isn't /bin a symlink to /usr/bin anyway? Kind of shoots down the old reasoning that /bin has the absolute minimum to get a broken system fixed, and /usr/bin has everything else.
One of our supercomputing clusters does run with a shared / and a separate /var that is local to the machine. /var is local to the disk because the nodes are stateless but we would like a persistence store for things like logs.
