The first honorable mention is by Ganger and Patt, on metadata update performance in file systems. They propose approaches for ordering on-disk updates of file system structures so that the information stays consistent (so you don't have to run fsck) without changes to the data structures, and improving performance at the same time. Finally, Kotz's paper on disk-directed I/O proposes a way to structure storage for multiprocessor systems so that data is moved in parallel from disks to application memory with little control overhead. This is accomplished by multicasting a request to the disk servers, and letting them control the transfer.
I hope you enjoy these papers as much as I did.