I have known about this game for a very long time, but I had always put off playing it because I understood it to be fairly difficult in terms of puzzles, and this impression was reinforced by a few sporadic attempts to get the ball rolling. That was something of a misunderstanding -- there's really only one puzzle, and though it is tricky it's not difficult in the sense of being hard to intuit what must be done; you have a definite objective from the start, and the scope of action is readily discerned.
All Things Devours is structurally composed of just that single puzzle; the provided story is the absolute minimum required to sketch out a motivation. I say that with admiration for the author's fine judgment of what that minimum was, because the game does a great job of providing just enough motivation to keep going even as the perceived scale of the puzzle's central problem keeps expanding in the face of early experimentation.
This game earns a lot of respect from me on a technical basis, especially for the elegant implementation of the core puzzle mechanic. (Side note: thank you to author half sick of shadows for releasing the source code!) However, the constraints of the puzzle feel somewhat artificial, and I'm not sure that I can put the label "fun" on a game that required putting together a spreadsheet to solve, even though I certainly did get satisfaction from the process.
One thing not mentioned in other reviews is the non-standard handling of the passage of time in this game. The author basically hotwired the Inform 6 Standard Library to be able to track time in five-second increments. As I recall, there are a few actions that require more than this span of time to accomplish; it would have been interesting to see this subsystem expanded to require variable lengths of time for different kinds of standard action, but that would have complicated both coding and gameplay in ways that would probably not be desirable. Although the overall impact is minor, one thing that is not minor is the effect on the player: Several reviewers note the way that the game instills a sense of urgency, and the extremely granular handling of the in-game clock does much to create and reinforce this feeling.
Another interesting sidebar is the code which handles duplicates of objects. These are not true duplicates in terms of dynamically created and destroyed memory allocations, but the subsystem is a remarkably clean way to handle the complexity introduced by allowing objects to be taken back in time. If you're interested in creating a strict time travel game with model of the process that presumes a single, unified and unalterable timeline, then you should definitely study how author Toby Ord handled the associated problems.
The game has been criticized for the fact that the story told by a successful playthrough depends on a superhuman level of foresight on the part of the player character, who finds that her original plan to sabotage her research won't work and must devise an alternate plan in a matter of minutes. I'm not sure how much that matters, since the real motivation for playing the game is the "because it's there" factor of it being a well-designed pure logic puzzle. I skipped the hard mode because it looked like the same type of exercise, only more so -- a little more story might have motivated me to try it.
If you like time travel stories, then this game will stand out as being one of the better "hard SF" implementations of the concept. I definitely recommend it for sci-fi fans and those who enjoy well-implemented puzzles that are well-grounded in the story. The technique used to implement the core restriction is also worth studying; it's a practically fool-proof method of handling paradox detection.