I imagine the increased surface area of the worm drive may cause more loss to friction than desirable.

Possibly, its less the reduction, more the "stroke" of the vibration (ie how far the tuning fork moves backwards and forwards.)

I think for a large tuning fork, you _could_ use normal watch gears. However I don't have the skill to try that, yet.

That makes sense; the backlash needs to be much less than the tuning fork travel, right?

Backlash isn't probably that much of a problem, the orignal movement has a locking "pall" that stops backlash: https://youtu.be/CPS7aNCAwAA?t=161

scaling it up you'd get something like this: https://www.youtube.com/watch?v=u48n-jvo5N0

to make it practical on a tuning fork, that movement of the forks would need to correspond to the tooth size of the wheel. For a large tuning fork, its probably in the order of mm, so not beyond home shop manufacture

A ratchet and pawl limits the backlash to the distance between the teeth on the ratchet gear; think of it this way; if it were to turn less than the radial distance between the teeth on the ratchet gear, backlash could still happen; similarly in regular operation, it turns the ratchet gear slightly more than that distance, and there is backlash until the pawl engages.