I can imagine that machines that used normal user-supplied decks might
have mono input, but surely even then one could find a suitable portable device with > stereo in, or even use a nice Akai or something.
Yes, almost all consumer tape recorders (except for expensive component stereo decks) had mono inputs.
The only reason for using cassette tape was to lower the cost of the computer, and being able to use a recorder that the customer already had was cheapest of the lot. Back then this was vitally important for breaking into the consumer market, because disk drives and controllers were hideously expensive. It made no sense to produce an interface for expensive stereo recorders because that wasn't the target market.
These weren't exactly the most complex circuits, I can't believe it
would add anything material to the cost of the AD/DA circuitry at
Look at the tape interface circuit of a Sinclair ZX81 or Spectrum - do you see any AD/DA converter? These computers were designed to use the smallest number of components they could get away with, to keep costs as low as possible.
...why not use FSK at several frequencies? Looking over the standards
I see frequencies from around 600 to 6000 being used, often at one end
or the other. For instance, Atari used relatively high-end frequencies
at 3995 and 5327 Hz, while KCS was 1200 and 2400. This implies one has
safe usage from 1200 to 5300 at a minimum. So why not both, or
Conventional FSK is actually not a good choice for audio cassette recorders (I suspect it was only popular because engineers were trained to believe that's how you do it). Atari's use of high frequencies is bad, and any system that tries to generate sine waves or filter frequencies is counterproductive. To use multiple frequency bands at once you would need to separate them with filters, and then distortion and noise become even more of a problem.
Better schemes ignore individual frequencies and just time transitions, realizing that the data is recorded on the tape as flux transitions which become a series of spikes when played back. They don't try to resolve high frequencies which are attenuated by head azimuth alignment and tape-head gap, and they take into account the frequency changes and random jitter caused by motor speed variations and tape stretching.
And those with digital interfaces, like the Atari and Commodore decks,
one could send data to the deck at a higher rate
The CPU in many retro computers is fast enough to encode and decode modified FSK at much higher rates than a consumer tape recorder can handle. Project O.T.L.A. creates audio files which load on a ZX Spectrum, Amstrad CPC or MSX at 12600 bps, but only when played by a device which has sufficient frequency stability and linearity to reproduce the signal accurately.
The problem with consumer tape recorders isn't just bandwidth, but also motor speed variations, phase noise, distortion, dropouts etc. Using a mono head or both sides of a stereo head improves signal-to-noise ratio and reduces dropouts compared to using one side of a stereo head. Using both sides of a stereo head independently would make it much worse. Cassette stereo is an awful format!