Is this because scanner pixels are rectangular?
No, the sensors are square. (Well, actually, they are probably circular.)
There is a linear strip of sensors that have a fixed dimensional relationship to each other. So you cannot "oversample" with such sensors. (Well, I guess you
could shift the sensor strip in its long dimension by half a pixel, a la Olympus's "High Resolution" mode, but I'm not aware of any scanner that does such a thing.)
So that's one axis. The other axis is controlled by a stepper motor that either moves the sensor strip past the item being scanned, or moves the item being scanned past the sensor strip. This is where sub-sampling can happen, as the stepper motor could move things at less than the natural pixel pitch of the sensor strip.
This is fraught with problems, and its not clear how sophisticated scanners are in this regard. The sensor array has a fixed "aperture" that overlaps when sub-sampling.
Simple sub-sampling results in no effective increase in resolution at all! The sub-sampled result will have more pixels, sure, but they will be fuzzy, and generally no better than up-sampling in Your Favourite Photo Editor™.
I haven't looked at the scanner market in some time, and so could be wrong about this, but I don't think the super-high-quality scanner market is big enough to provide the R&D budgets necessary to do the sort of magic Olympus does with High Resolution mode. So I don't think any of those claims of 2,400 spi really hold any water, except in marketing literature.
So I'd welcome updated information anyone might have regarding the reality of resolution on scanners that claim to have more than 600 spi resolution in one dimension. (I believe 600 spi is the maximum pitch available in sensor strips.)
So how do drum scanners achieve 5,400 spi? They don't have a strip of sensors with a fixed dimensional relationship to each other. Rather, they have a tightly focused spot of light that is picked up by an extremely sensitive photomultiplier tube. But it takes a lot of power to squirt enough light through a seven-micron aperture and still be picked up by a PMT! So the highest-resolution drum scanners used some 500 watts of light to do so! They were beasts, and my "compact" Color Getter Falcon (with a 500 watt xenon carbon-arc light source) weighed 100kg (220 lbs) and filled a 48"x30" table top!