There are two ways that explosions can happen. In a LARP where the action is driven by plots created by the GM, it's an escalation problem. The GM is continually having to come up with plots that will interest and excite the players. The longer the game runs, the more the players have seen, so things have to get more intense to remain comparatively exciting. Threats get more dire, strange happenings get more bizarre, the stakes get higher, and around the time when the players are saving the world from an incursion of interdimensional zombie Santas for the third time -- this time while also trying to woo the Martian monkey-bats' diplomatic envoy and hunt down a rogue nuclear bomb at the library -- their disbelief suspenders just snap and the whole thing comes crasing down. (That, or one session they actually fail to save the world, and then what?)
One of the strategies to avoid this is to put the GM contributions on the back burner and focus more on player-driven action. This BYOP (Bring-Your-Own-Plot) approach has a number of things going for it: not only does it relieve the GM of trying to come up with something that will interest all 30 players every freakin' month (since each player is pretty much guaranteed to be interested in whatever plot he's invented for himself), it also accommodates a disparity of relative power levels, since a brand new player is just as able to come up with good challenge for his character as an old hand whose character has years of buildup.
The problem with BYOP is that any time the players are allowed to affect the game world in a significant way, someone will eventually break it. Somebody will come up with a plot for themselves that involves toppling world governments, or inventing some radically disruptive technology, or collapsing the barrier between dimensions, or, even if they stick to mundane activity, a slow but steady accumulation of changes from the real world that eventually make the game world so different from normal reality that it's impossible to keep track of everything. (This is also a problem in GM-driven games, but often doesn't crop up before the escalation spiral starts to kick in.)
Which puts the GM in a bind: either prevent the players from doing anything meaningful and significant, making everything static and dull, or let the game self-destruct after a certain amount of time.
To introduce my idea for solving this conundrum, I have to tell you about the card game Mythos.
Lo these many years ago, back in the paleolithic era of gaming when CCGs were still new, my friend Chris discovered Mythos, the CCG based on Lovecraft's Cthulhu mythos. He and I bought a lot of cards and built lots of decks (because that's how we did CCGs: whoever was into it would buy all the cards, and everyone else would come and play), and we'd have big eight-person games at the coffee house.
Now, there are two ways to end the game in Mythos: either someone plays the right cards to get a total of 20 victory points and win, or someone gets taken out by losing all their sanity points, in which case the game ends immediately and whoever had the most victory points at the time is the winner. As a group, we eventually came to the conclusion that winning by taking someone out was much less satisfying for everyone. It's a cheap victory, it introduces kingmaker politics, and it dramatically restricts the kinds of deck that are viable if you always have to worry about having enough defense. So we came to a gentlemen's agreement not to play that way. You can use the threat of losing sanity to slow someone down, but taking them out? Not sporting.
And thus was born the moral victory: whenever someone was in a position to take the cheap win, they would point it out, receive the appropriate recognition from everyone else, and then refrain from doing so and focus on getting a serious victory, because we all had more fun that way. (The idea has spread somewhat since then. These days, we frequently award moral victory points to people when they are unfairly robbed of winning the game by a bad roll of the dice or the like.)
It's a kind of counting coup, and I think that's the solution to the exploding BYOP LARP: build some type of corrective system into the game world or the premise of the game that lets players get credit for being in a position to do something clever and world-breaking and not actually doing it.
"Credit" can be any number of things, depending on the game. It might be that the player's superiors give him a boost in status as recognition of his accomplishment. It might be that the results are announced to all players, and then time is rewound by a month and things happen differently. It might be nothing more than a "well done!" from the GM and the player's knowledge that he broke the game.
All of these approaches could work, as long as it's established and agreed upon at the beginning. It could even be that counting coup is the whole point of the game, and the players' goal is to do it as many times as possible. It might be interesting to have the other players decide what is and isn't world-breaking. Players might literally have a coup score of some kind, either secret or publicized.
The key is in recognizing that game-world breakage is likely to happen and that it's not a sign of things going wrong, it's a sign that a player has been clever and organized and has pulled off a cunning scheme. It's an accomplishment. It's just an accomplishment that can't be allowed to actually happen, for the sake of everyone who is enjoying the game and doesn't want it to end just yet.
I think it would work spectacularly well in a game like Cognoscenti, where everyone is playing the high-powered Secret Masters who are out to reshape the world, but I'm trying not to get carried away with redesigning someone else's work...