Wow, that would be an amazing and ambitious project! And you are absolutely right that it is the presentation that is the easy part.

Hmm, I guess some questions to ask to get the ball rolling would be:

Here's an (informal) precedent from the 50s:
http://themicrofoundry.com/other/trends.gif

This type of thing can be extremely useful. Please do it!
But yes, as Dana has sort of alluded to you need to focus
it quite intelligently and pragmatically first.

hhp

I was thinking of selecting a range of 20 or so publications that actaully span a decent length opf time. Quite a wide range too, trade mags, professional mags as well as pop mags. Then, go through each, page by page and write down everything. I'm not sure if that's a good means of getting a sample, but it seems pretty good right now. Any ideas about how to effectively hone a sample group? I was envisaging this being a group effort in terms of the actuall analysis, it would take AGES!!
It would be worth perhaps making the spread of info display not only type, but when and where (country/media) too. This means film, swing tags etc. To get the ball rolling, a focus in the interim on magazines might be the go? It has to start somewhere...

If you go with sampling an array of magazines over a time period, the next question is what will count as an instance of one font. Magazines will use a specific font for text over the course of a few years, as part of its style, but advertisements for a product may use an array of fonts, depending on the campaign. Would you count each year that a magazine uses one font as an instance of usage, even though it's one continuous use?

I suppose each issue would need to be counted in its own right. That is a good question though.

To get a good sample of magazines, maybe compile the magazine subscriptions of public libraries. Since libraries serve a public readership, which is anybody and everybody, their subscriptions are usually pretty well rounded. This could be a starting point, but I would imagine that you'd of course have to add to the list.

Also, when you collect data for font usage, how would you compensate for a font that is really visible versus not very visible? Like a font for newspaper headlines versus one for photo credits.

Good idea! In addition to Danas first comment I suggest that the survey should also be split into groups of fonts – e.g. serifs/sans etc. - for ease of use both in data accumulation and in readibility of the "output". There are several web pages that give the front pages of daily newspapers, I once categorized the ~400 "logo heads" (i.e. the newspaper titles) for a day into several groups of typefaces (sans, serif, typewriter/egyptienne, blackletter, non-latin, plus Caps in those categories) and country/continent. It was somewhat restricted to the pictures I could accumulate through automated downloading w/ wget, but it showed quite a lot of different uses of type in different regions. It was a heck of work, and I lost it after not making a backup copy.
I imagine one could combine such an approach with the question of unique type faces and think that such a survey could help quite a lot both in identifying typefaces from a certain period of time and in designing new stuff.
One more question that comes to my mind: What about combinations of typefaces? I think that these contribute to the appearance of a designed product and should not be forgotten (although that would further complicate the survey).

Ohh man. Losing that work must have been pure heartache.