There's some precedent for your question.
First and most obvious is the Matrix, which touches on the idea of suffering in a simulation. But you're asking about a being (an AI) that, unlike the masters of the Matrix, has some sense of morality.
That question, famously, of how a god-like being ought to behave, has been explored a little already.
Going back in American media: the 1939 movie version of the Wizard of Oz puts a human in a god-like position, where the suffering he can cause contrasts with his humble Omaha roots. In the Twilight Zone episode The Little People (aired 1962), the morals are made pretty clear. The "god" is human and acts exactly like your "dungeon master".
This theme of using power to cause or alleviate suffering is also a constant theme in more narrow sci-fi genres such as the comic books of the "Golden Age", (WWII and post-war). It can be traced back further in the English language, through Victorian supernatural where the "tyrant" choice was pretty common, back to Gulliver's Travels in the 18th century.
The VR world
What you seem to be adding is an AI god, who may be better at planning than the humans in the above examples. And not just power over the inhabitants, but also the ability to reshape or create them.
To play up this angle I would suggest:
- a mutable, changeable VR world: a simulation where features can be changed on the fly, to best explore your question.
- an AI who, in terms of personality / morality, is essentially human. Otherwise, their goal, in terms of morality, might be unclear. On the other hand, the enhanced intelligence of an AI protagonist could help you better explore the consequences of their designs. An AI could predict / recount future events and chains of causation, like exactly what might happen when they start lobotomizing everyone, in a way that a human can't.
Pitfalls in creating this world come from dealing with a setting / timeline that changes so much, and keeping the plot straight as your AI explores the different options.
- What happens to Chekov's gun when the world resets?
- How do you keep characters relatable when their brains are putty for an AI?
One standard way to get around this would be to have a low-powered, outsider human antagonist, who explores the world the AI creates. A very recent example of this is the highly mutable world of the Nightbirds series by husband-and-wife authors Tracy Hickman and Laura Hickman. Using this naive perspective, you can build a lot of tension, and, once you start rearranging the scenery, you can get in some great imagery (check out the excerpt from Unhonoured). You may have to use the Amnesiac Hero trope, as in Nightbirds, to gently introduce the reader/player to your ever-changing VR world, where the simulated beings suffer daily and are "helicoptered" in turn. As for "what is more immoral": how that turns out is up to you.