It might be useful to define the term demo first. To my understanding demo in the questions context means a in functional (level, etc.) restricted game meant to be given to end customers so they can test the game and buy the real copy later on (lets call it Demotype A), versus demo as in a not for resale demonstartion of the whole (or at least in stage of development whole) game given out to potential sales partners, development partners, or journalists for pre-gold reporting (lets call it DemoType B; *1).
As of this, the mentioned example is only a a demo of the later kind (Type B). For this kind per unit investment is rather irelevant, as doing a presentation where it gets used costs way more than the hardware needed to produce one (or a few) - which isn't cheap either (*2). For this kind of usage 'only' a development snapshoot (test version) is needed, as it's not intended for use by end users (*3). Everyone targeted will be aware that it's a preliminary, potential untested and buggy version. A restriction of features is not neccessary for this audience - saving to put additional time constrains to already stressed development (Is there any game development without?).
Producing demos of the first kind (Type A) in contrast must come with a cost comparable to other kinds of advertisement. Maybe a bit higher as it is more targeted, but not much. Making a CD, even with a run of just a few thousand, was already way below 50 cent per unit, including a printed cover. And with higher volume the cost dropped even further, making it feasible to use it as advertisement.
On the other hand, making a cartridge involves not only the standard parts like PCB, plasic hull and sales cover (which in itself already go past the 1 USD mark in the 1990s) but also making a ROM. Doing a ROM has a base price of >20 grand for mask and run cost plus per chip cost. Switching to programmable technologies only saves for very small runs, as it only removes (most of) the setup cost, but adds higher per unit cost - not just because the devices used (PROMs) are more expensive than similar sized ROMs, but due the handling effort in programming them.
So handing out a demo-CD will be not much cost to start with, demo-ROMs do have a hefty initial price tag. And while both decrease over volume, the starting cost for a cardridge are much higher than for a CD - and even in high volume will not reach a region even remote acceptable.
Bottom line: Way too expensive.
*1 - Supercat's 'answer' reminds me that there is another possible meaning of 'demo', not targeted by this question: Demo as an art form defind and nurtures by the demoscene. These demos are written for competition or fun, showing off the capabilities of machines and their programmers. While not realy related to the question asked, some of these demos get distributed as cardridges in (very) small production runs, but even for module based consoles, the usual distribution form is via archives on the net - an usually free to be copied anyway.
*2 - Keep in mind, we are talking cost for a company, not just cost of parts as some hobbyist may count. Each additional hour developers have to spend to prepare a snapshot, the time this snaopshot gets moved into an (E)PROM, someone dies a label and explains the dos and don'ts to whoever is doing the presentation costs hundrets of dollars, making the material bill cost irrelevant.
*3 - R.. Asked about using such demos for potential customers in stores or at trade shows. In such situations a restricted version does not make much sense. When a game hits stores, a regular cardridge will do the same, so not worth any investment.
Similar for trade shows after the game got published. Now for shows prior to release, game developers are rather relucant to let anyone in the general public play the game. Any bad user experiance due being unfinished will for sure result in a bad reputation of a game (happened almost always when developers tried that route). A marketing desaster no developer or even less publisher wants to create. Not to mention that, when going gold closes in, developers won't have much time to curate, test and debug a special demo version anyway.
So, while the hardware may not be a big deal for such a small number of demo cardridges, their implications are way too dangerous to even think about.
Last but not least, for pure in-store displays a recorded play video (real VHS video in the 80s and 90s :)) will be way more apropriate to shoow off features without risking bugs to damage the rep and without all the costs associated with it. The one playing the demo will be someone knowing to only use features that are already finished (and intended to be shown) - basicly replacing the whole eford to code the demo parts at all.