Dungeon Master should get a slow, time-based income (to give an urgency to the player's actions), supplemented by income from player actions (a very nice idea, probably the most original one Yahtzee had there). Dead monster value should be re-cycled through a pool that supplements the time-based income, adding to each tick until their value has been returned - and may also be more quickly transferred by particular player actions.
That would give the dungeon master an army that slowly grows in value (whether that's number or individual strength).
Spawning could be done through fixed points throughout the level (see Dead Space's air vents), anywhere outside of line of sight of the players, through some sort of portal/rise from the ground/other animation that allows it to take place anywhere with only a delay and graphics allowing the victims to respond or some combination of those traits.
These methods would allow for flexibility in unleashing the minions, potentially allowing unlimited variation in games.
The player objectives could be as simple as navigate a maze, as in Yahtzee's idea, or your more typical involved game's route of a series of buttons/events to open up access through a relatively linear level. There's limitless potential in making a map, but the general gist should be worked out so that decisions about the impact of player actions on the DM's income can be assessed with regard to game balance and length.
Finally, the players themselves need some sort of progression, especially as the described nature of the DM's army and income mean that he only gets stronger over the course of the game. Players will need to be able to gain steadily more powerful weaponry and/or abilities to survive, with a balance needed between taking down a steady stream of lesser foes to periodic hoards of them or packs of more powerful enemies, depending on the DM's style of play.
Zombie Master was a fun mod, but the 3D nature of the maps actually worked against it in several cases, with people exploiting awkward ledges and the suck like to make them invulnerable to the zombies. Add in a range of qualities of maps, with some of them lacking focus and direction, plus the generally shaky nature of early development stage HL2 mods, and it kinda struggled. Shame that it was abandoned as the premise was pretty good.
To be honest, if SC2's third person mode didn't lag so much I'd really enjoy knocking a game like this together over the Summer in the map editor, but it does lag, and this really wouldn't work as well using normal strategy controls. Maybe a top-down shooter... I wonder if the control lag's just as bad is the camera's up there... hmmm...
