What I did (Datalight, Zortech) was make large disk I/O buffers. It gave huge boosts to speed, and very few people cottoned on to that :-)
It was a technique I learned from Shal Farley.
Fewer round trips to the operating system almost always speeds things up, too, especially if it involves a context switch.
Things got even better when the switch to hard disks was underway. Hard disks had bigger sector sizes, but the buffer size sometimes remained at 512. This meant the driver had to read the sector, fold in the 512, then write it back out. Oops!
Later on, demand paged virtual memory and memory caching also favored big buffers.
It is true however that sequential access to hdds and floppies was always much faster than random access or even sequential in small buffers simply because the drive does not need to wait for the platter rotation to reach the right location again for each sector written but rather just write them all in one go.