It's frankly boring that 20 years later we still feel attracted to discuss any of this stuff. It's not like the machines or the code running on them have fundamentally changed in the interim..
Another example, I might be wrong here though, is the ooold disk handling code in BSD, where you had to specify cylinders, heads and other things, and the OS tried to optimize performance based on closeness. See here for more info: http://www.cdf.toronto.edu/~csc369h/fall/lectures/w10/DiskIO...
Today, I imagine most I/O interfaces use a large block index instead of cyl/disk/head and other ancient specifications. All the optimizations about head and so on is device dependent and is best handled by the disk itself.
Imagine the feeling of being able to delete old code in the disk subsystem and get a cleaner and more abstract view of the disk.