It's far older than the 90s, but that's a product largely of what old papers have been digitized, and minorly on the terminology—I think you'll find some papers that say 'cognitive workload.'
I recently fixed up an old bit of scientific apparatus, which was used to measure the cognitive workload of various aircraft cockpit designs. It used a method called visual occlusion, which you can guess the rest of—the pilots would sit in trainers with a set of glasses, that had spring-loaded solenoid flaps to block the pilots' vision selectively. The open time as well as the frequency of the system was experimentally varied until the pilot could fly the plane safely, for given conditions. The vacuum tube circuit that I repaired had a date stamp from the mid seventies but this technique was already more than a decade old by then. The same technique is now used (with liquid crystal shutters) to evaluate interfaces in cars for drivers, maybe a variant would be useful for certain application areas of software (modulating the screen, perhaps)?
I poked around myself and the earliest use of the term I can find is in an article from 1968:
Pupil size and problem solving
JL Bradshaw
The Quarterly Journal of Experimental Psychology, 1968
Of course when I tried to read the text of the article, I got a message saying
Sorry, you do not have access to this article.
How to gain access:
Recommend to your librarian that your institution
purchase access to this publication.
If you search that book for "cognitive load is defined" you get the snippet:
> "Cognitive load" is defined loosely as the amount of mental strain put on a person during the performance of some task, often at least partially due to the constraints placed on the performance of that task...
The related term "mental load" seems to be a few years older, coming mostly from precursors to HCI, such as industrial ergonomics. Here's a use from 1961, also paywalled, but with the abstract available: http://www.tandfonline.com/doi/abs/10.1080/00140136108930502
"Pupil size changes were monitored during the solution of various types of problems. A number of solution and response strategies were required of the subject. There was strong confirmation of the theory that this autonomic index can provide a sensitive measure of the fluctuating levels of attention and arousal, which are associated with the various aspects of information processing and response."
In relation to the web, a quick Google Scholar search brings up lots of papers from the 1990s: http://scholar.google.com/scholar?q=%22cognitive+overhead%22...