The three-act structure has been attempted and bungled in some games, but that's not the same as it not being a viable option for game design. It's a tool, it has its uses.
For example, the structure as represented in one of my favorite games ever, THIEF: THE DARK PROJECT (with spoilers):
ACT I: We are introduced to Garrett the thief, the medieval-industrial game world, and the core components of stealth, detection, and action that define its gameplay. The game riffs on this core model with a few largely self-contained levels that build the character and atmosphere. We rob Bafford, free Cutty, and plunder the Bonehoard.
ACT II: Victoria shows up and hires Garrett (i.e. we the player) to obtain The Eye, which it turns out will require more than one mission to accomplish. We even get within site of the prize in one level and have to leave and prepare, over the course of 2 or more levels (depending on the edition of the game you've got), to come back and complete that mission. This confounds our expectations of how game levels typically interact, and feels a little non-linear, even though it's completely linear. Eventually we gather the necessary MacGuffins and capture the prize: The Eye. But when we turn it into our patron, he turns on us, reveals himself as an evil god, steals our eyeball (!) and leaves us to die.
ACT III: Garrett spends a level escaping, continues shifting and altering loyalties established in the first two Acts, and embarks on climactic revenge missions against the revealed arch-villain. It's more than a boss fight, and establishes a new tone for these final missions that wasn't present before.
We get three acts divided over something like a dozen levels, drawing on classic and filmic narrative structure and traditional level-based game design at the same time to create an experience that allows a ton of player freedom within the individual story beats of an otherwise fixed tale of crime, betrayal, and revenge. THIEF didn't surrender game for story or vice versa.
So I don't think it's accurate to say the structure *doesn't* translate to games. It's more accurate to say it often hasn't or that it doesn't translate well to all genres of games. It's absolutely accurate to say that the formula is fallible -- confounding expectations can be vital to good game design and good storytelling. Some great games tell terribly tired stories and are none the worse for it, while others tell enjoyable stories through solid but not groundbreaking gameplay. These are both fine outcomes, provided they are not the *only* kinds of games we're making.