Imagine trying to design a calculator that isn't allowed to display the digit '8' on its LCD display.
In software, the implementation sees if the LCD is being set to a value that contains '8' in binary and if so, ignores the input.
In hardware, you have to have a camera that is pointed at the LCD, and a hardware OCR implementation that looks for the 8, and then a circuit that when the OCR implementation sees an 8, turns off the display.
1. If you have a single radio unit, you can easily do the "software" implementation where you look at what you're going to be broadcasting and cut it off if it's not legal. The problem comes when you have separate radio and amplifiers - the amplifier has no idea what whether you're feeding it 10M or CB frequncies. It will amplify either just fine, so the FCC really prefers you to build in a "hardware" type implementation there (a frequency analyzer that figures out what's going into it). It's just expensive to do so - like if you didn't already have a digital amp, you might easily add 1/4 or 1/2 to the cost just to get some dumb logic built in. And you can still probably get around it if you really try.
2. There is a disproportionate amount of cheap gear and old gear in circulation. We didn't get "smart" rigs until the late 90s at a minimum, and even so they are vastly outnumbered by stuff from the glory days of the 50s, 60s, and 70s. There has always been a relatively small amount of high-power stuff in circulation, and there's even less smart high-power stuff. A lot of the modern kit-built stuff like the Elecraft K2 focuses on QRP (effective operation on low power) instead of crazy high output.
3. This manifests as a CRAZY price curve. It's relatively easy to get an old radio from the 70s that'll do 50 or 100W (especially after factoring in weak, aging tubes, etc). A modern mini-rig that will do 25W or something is also relatively affordable. But if you want something new, powerful, or new AND powerful you can easily jump into the thousands of dollars. For the cost of a modern rig that'll do any given power level, you can afford an older rig and an amp that will put out an equally good signal with substantially more power. Or with a better antenna. And so on.
4. There's nothing wrong with the old gear and things like tubes can be fixed. If you trash your new wünder-rig it's garbage. So given that amateurs are crafty people (the proto-hacker, really) there's a strong incentive to make the old or lower-power stuff work, because it's repairable when it breaks.
Why is there significant expense in putting a ROM chip between the radio and the amplifier that cuts out certain frequencies? I can see where that's not as simple as mandating software that is simply incapable of generating bad frequencies in the first place, but it intuitively strikes me as the same sort of embedded programming that goes into, say, a microwave oven--I can see it adding some cost, but not so much that it would be unfeasible in e.g. an home wifi router. What am I not understanding?
Adding a protection device merely shifts the problem (being discussed) from the radio to the protection device.
It achieves exactly nothing.
An example of the former is "homemade AM radio", where you wrap a lot of wire in a coil, and changing where you complete the the circuit along the coil allows you to "tune into" different frequencies.
An example of the latter is the cantenna, a directional antenna that can be made from a Pringles can.
The list of blocked frequencies changes occasionally (like time zone definitions); if you can't update the "illegal frequency blocking" chip then it becomes illegal to use your otherwise-perfectly-good product once the rules change. Plus, it depends on what country you're in, so ideally if you move (or just go on a vacation) to another countries you don't want to have to throw away your old router/smartphone/smartwatch and buy new ones in the other country.