In my previous post re LAN traffic at the University of Venice, I just realized a mistake I'd made. (Funny how time plays with your memory.) The entire university was NOT one giant collision domain! When I first started working there the university was spread-out in temporary offices all around the city; a new campus was under construction. So, many of the departments were linked by routers and ISDN lines. The problematic LAN was limited to the departments/buildings that were directly linked to the main computer center. Still, this was enough users that those departments' Novell servers showed enormous statistics for lost packets.
And, of course, my primary point still remains -- that switches rendered moot the old argument about NIC design.