The need for logical stopping points.

This is probably stupid and unnecessary to explain, since "logical" implies that it's clear why you'd stop. I'm going to attempt it anyway, instead of leaping straight into the discussion of replayability. This is all tangentially or fundamentally or necessarily tied to a conversation about pacing, so I bet (if, as I am, I'm appending this sentence after finishing the post, so as not to completely surprise you with where I end!) we might talk about that too.

By this point in the conversation, your brain has already jumped straight to save points. Though originally driven by technical limitation, the artificially constructed save point still exists and facilitates session length. Save points are a specific subset of checkpoints (I guess by some definition you could say it goes the other way around), which are markers in a game's timeline or landscape before which your progress will never be reset, subject to optional conditions like a try counter (lives/continues). Save points are sometimes thought of as persistent checkpoints, but since plenty of games have checkpoints that are persistent anyway, there's not much reason to distinguish between the two. So checkpoints are the easiest way to define a session.

Traditional JRPGs (Ezio mentioned Final Fantasy IX in yesterday's comments) demonstrate the most rote implementation of this. By placing saves in safe locations like towns or the beginning of a dungeon, they communicate that a session is meant to begin and end in safety, with the middle composed of danger/exploration. This sets up a basic narrative formula: Act I contains the collection of information and development of plot, Act II is the search for the objective, and Act III is the peaking action and climax (e.g. boss battle). Games patterned after this architecture are thus stories within stories, each session analogous to a film or play. Whether you're playing something like Metroid Prime or Zelda, you're getting dozens of these.

More common in Western and contemporary gaming is the notion of save-anywhere, or save states so frequent that they are essentially continuous (see Basically Everything in this day and age of auto-saves). This has led to the abstraction of narrative chunks from saving; instead of a play session with a beginning-middle-end, we have a quest or level. This offers a great freedom in the portioning of content, along with the dangerous temptation of decompression. As a creator, it's too easy to be content with a narrative divided into quests, each quest instilled with purpose but not carefully measured.

Long adventures managed by a floating arrow or quest log reminding us what we're doing also remind us that we don't even care enough about the destination even to remember it. Current WRPGs like Fallout 3 and Oblivion/Skyrim try to excuse this by throwing lots of random shit all over your path. Hey, you were on your way to go rescue a book from a dungeon, but look, ants are attacking this town! Better stop to shoot humanity's way back to survival of the fittest! Whoa, and some kid is hanging about looking for his family! Better stop shooting ants to find his brother-in-law, before he has to spend Christmas all alone! What you get is a random-ass structure where the build-up from the first ten minutes of your play session may be supplanted by a completely different climax, and then two weeks later when you get back to saving that book, you don't even care what it's about enough to read the dialogue. The game pummels you with starting points without naturally pairing them with endings. I'm not dismissing or condemning spontaneous or uh, what do they call it these days, I think "emergent" gameplay (Arkham City is a successful example), but if you ignore the constraints of playtime and don't give the player somewhere to stop, they're going to turn off the console after forty-five minutes and take nothing away from it.

Equally precarious are tiny quests so small as not to demand any justification of their own, with the expectation that the player will plow through a half dozen every time he sits down. There are stopping points everywhere! You can determine your playtime however you want, be it five minutes or two hours! This is a common strategy for open-world random shit games, the "GTA with ____" class like Prototype and Red Dead Redemption, also rearing its semi-ugly head in post Super Mario 64 collectathon platformers. This can derail the player with meaningless forgettable content and over-stimulate them with variety. Prototype is definitely a case where tiny objectives make the game feel thinner than it should (Prototype 2 works better, though is still flawed), while Super Mario Galaxy (not to mention the rest of the post-64 series) and XCOM get away with it by being addictive as fuck. Animal Crossing sits somewhere in between, where some micro-missions are outrageously pointless and others build just enough toward a goal. But addiction is part of tomorrow's continuation into the subject of replay.

The foremost concern in pacing needs to be the length of time the player is intended to continuously play the game. Platform and genre provide restrictions on this, but at the end of the day it's dictated by the game itself. Establish a session with concrete aims, rewards, and progress, then bookend it with logical stopping points. Impact and memorability can be developed over a length of time and countless sittings, but only as the sum of parts. That means no part should count for nothing.