When a transactions is created in the one cluster, it may not be transfered quickly enough to the other cluster, where the next block happens to be found. Obviously, zero confirmation transactions will be of less use.
Due to the same starved link problem, a block may not propagate to a miner in the other cluster quickly enough, leading to a higher orphan rate, and may be to orphan chains of some length. That means the six confirmation safe level may have to be extended. Six confirmations have always been an advice anyway, for each confirmation the confidence increases.
In such a scenario, one cluster may prove to be the leading mining arena, and any miner must make sure that they are well connected to that cluster.
But will it take down bitcoin? No.
Moar nonsense from someone who apparently has insufficient experience in software engineering. Sorry to make this personal (especially with someone who has expressed some appreciation for my input on the forum yet with reservations about my etiquette), but some of you guys have no self-restraint. Some of you (and you know who you are) go on and on and on posting about technical issues without even the slightest bit of reticence nor shame that you might be totally-fucking-wrong. The reason this thread is not in the Development & Technical Discussion board, is because many of the posts in this thread would likely be deleted by the moderator there to prevent the promulgation of misinformation.
Hey n00bs, a starved link means it can never catch up with the transaction rate. Infinite confirmations won't help you. The two clusters will diverge as two forks with the BTC money supply effectively doubled if every HODLer can access both clusters.
The only way to deal with this situation is treat each of the clusters as pegged side chains of each other, then lock the value for each coin on one of the clusters (which is something plausible on a starved link such as a short wave radio feed). This is yet another reason I am excited about Blockstream's work.