Magnetrons are cheap, reliable, and work well enough. Their one and only job is to produce energy at about 2.4GHz, and to make a lot of it. You merely put in electricity and through the magic of geometry you get RF out.

A transistorized solution is, simply, inappropriate. The RF section alone will cost more than an entire microwave off the shelf, and plus that you need all the control circuitry and high speed crap to generate a 2.4GHz tone. Not to mention that putting kilowatts through silicon creates a much higher risk of catastrophic failure: transistors blowing out into a plume of plasma and such.

You're asking why we haven't reinvented the wheel. There's simply no improvement to be had.

You want high power RF in a given frequency band, you want a magnetron. It's just a lump of iron with some wires. You don't need any electronics. At all.

Because transistors for generating even very low microwaves like 2.4GHz are extremely expensive comparatively speaking, and don't produce much power. They're good, though, because you can produce very precisely tuned and modulated signals and very precisely controlled output powers - as long as they're less than a couple of watts.

A cavity magnetron is a block of metal with some holes drilled in, two bits of glass glued on, and all the air sucked out. They're hard to tune to exact frequencies and hard to regulate to exact powers, and modulation is as you've already discovered kind of limited to just turning them on and off - but they're extremely cheap to make, last a very long time, and require minimal support circuitry to generate double-digit kilowatts of RF.

You don't need to be cock on frequency to heat up a pie.