The BTC Guild owner could make a large payment to someone (say to buy a car). The car seller would wait 3-4 confirmations for assurance that the network has accepted it. But then, since the BTCGuild owner owns 50% or more power of the network, he/she could outpace the network after those 3-4 confirmations and essentially get the money back from his car (by double-spending).
Pools are only able to remain large as long as people trust them. We are at the mercy of our members in order to maintain our hash rates. A double-spend can be detected if anybody is looking at the work the pool serves (it wouldn't be working on the main blockchain, prevblockhash would be different). It can be *proven* quite shortly after (blockchain.info orphaned blocks page would show the pool doing this), at which point that pool is going to be dead shortly.
As a result, here's a little more to consider: There is nothing available for easy purchase with Bitcoin that is valuable enough to even remotely tempt a pool operator to do this. Anything that valuable would be a private deal, and you can be damn sure the person doing that deal wouldn't complete the sale til FAR more than 6 confirmations. The only method to generate the value that this could be worth it is via an exchange. Exchanges do not execute withdrawals in the $100k+ range automatically, so by the time this massive withdrawal is even considered, news would be out of the double spend and they'd manually block it from actually going through.
So for a pool to do a double spend...it's virtually impossible to do in a way where the payoff is worth killing off a viable business. The same is somewhat true of private mining companies: The attack would be publicly known and odds are would be blocked by the person they're trying to trade with in some way. The only difference there is the private mining company wouldn't also instantly lose their share of the network since they don't rely on the public to keep their hash rate strong.
