I'm glad my post got someone to ask some interesting questions.
Could D&D on the dice like this work? Has anyone tried it? Would it lead to more headaches (nonsensical monster encounters, overpowered treasure) or less (freedom from PCs trampling or otherwise screwing up plots and such?)
Yes, to a point. As others have said, you just painted a good picture of 1E. It's also a good picture of my gaming table, to a point.
Since my post generated the topic, I'll just talk a little bit about how I run my game and why I don't go further down the road of "procedural everything."
First though, I need to make the difference clear between the terms "procedural generation" and "random generation." PG using a chart and a d20 (or d100) to generate attributes of an entity (town, guild, person, whatever),
but the charts are not random. For my worlds, I write these charts and determine the odds of any given result. For 1E, Gary provided this service. It's very important to understand that you are making editorial decisions when you draw up your charts; what monsters are on it? what are the odds of meeting them? races? is "tudor mansion" a possible result? The world is not random; it's procedural. You (as the DM) don't know what's over the next hill any more than the PCs do, but you wrote the chart that gives the answer, so you're still responsible for the results.
As for my campaign, I like to draw maps (it's a hobby), but I also borrow liberally from history and present-day. I've got a map of 12th century France I've been dying to use, but I just haven't had a chance yet. What goes on the map though is procedural - towns, guilds, forms of government, local religions, etc. are all procedurally generated. I keep lists of NPC names (organized by gender and culture) and just go down the list of people as I need to drop of name.
PC: "Who's the Duke of Gheldaneth?"
Me:
<glances down>"Sucount Meros Ur."
I keep a person wiki to keep them all straight.
There's no meta-plot or story that I am seeking to "tell" to my players. I ask my players (at character generation) for their character's motives, preferences, interests and friends. I then (creatively, not procedurally) think of "bangs" to get the story rolling immediately. This is important because I want my players to actually care about the events that are unfolding. I need to link the events and what's going on to their stated interests so that they have a motive (other than payment).
I also then (creatively, not procedurally) decide a few things about the "bang" and why it happened. I have already created (procedurally) a world to inhabit, but events within the world are not random - there must be plausible cause and effect at the micro-scale, or players quickly lose interest.
As an example, I just started a new 4E Forgotten Realms campaign in Gheldaneth, the largest city of High Imaskar. Session 1 the PCs are sitting around a table at the Inn (they're all friends before play starts, I required this) and a messenger runs in, grabs one, and says "Your father's shop was just robbed in broad daylight! They didn't take any money from the till, but they grabbed something from the back room where your father keeps his most valuable possessions! He has asked for you to come immediately."
Bang. And we're off.
In my experience, there are certain things players will accept if procedurally generated, and others that will not be:
Things you can procedurally generate:
1. The motives and personal appearance of a given NPC they have never met before.
2. The monstrous inhabitants of a cave they've never been to before.
3. The political structure of a town they've not been to before.
4. Cash & non-magical treasure.
5. The race of a thief in a given large city.
etc.
Do you see the pattern? If the PCs have not encountered a given thing before, they'll believe anything within reason. As long as your charts only given reasonable answers, you're golden.
But your players expect cause-and-effect. If the main church of a theocratic city-state is devoted to a Lawful Good deity of justice, they just won't believe that city's mayor a Chaotic Evil mindflayer. That's an extreme example, but you get the idea. The more the world is fleshed out, the less often you can accept the answers your charts give you. Since they create path dependencies, the charts become less reliable every time you use them.