This was the approach IBM used in it's first RAMAC RAID where I think they had to buffer a whole cylinder but that was many generations ago
(my copy of the specs may not be exact): Buffering a whole cylinder, or a whole surface, of the RAMAC was no big deal. One hundred surfaces (52 platters, but not using bottom of bottommost nor top of topmost) totalling to 5 million 6 bit characters. That's 50,000 characters per surface. OR 50,000 characters per cylinder ("square geometry" :-)
On Fri, 15 Apr 2022, Paul Koning wrote:
"Was" as in "back in the day"? 50k characters would have been quiet a large memory in the 1950s. And for an I/O device, any kind of buffer is not necessarily all that useful.
50K would be bordering on extreme in the 1950s, and considered a LARGE buffer through the 1970s. Certainly wasn't practical with less than 16 bits of addressing.
What does make sense is a track buffer in drum memory machines, as found for example in the Dutch ARMAC, where the first implementation of the famous "shortest path" algorithm was first implemented.