From Satoshi's paper (
http://bitcoin.org/bitcoin.pdf):
If two nodes broadcast different versions of the next block simultaneously, some nodes may receive one or the other first. In that case, they work on the first one they received, but save the other branch in case it becomes longer. The tie will be broken when the next proofof-work is found and one branch becomes longer; the nodes that were working on the other branch will then switch to the longer one.
In the event of a perfect tie, 50% of the honest nodes would be working on one branch, and 50% on the other. That means an attacker with only 25% of the power would have a 50% chance of choosing the next block. The false block would break the tie, and all the honest nodes would now work away at the new longest chain (the one with the false block). An attacker might even be able to detect ties, and thus know when to try to attack.
The frequency ties, and how evenly computational power is split during a tie, probably depend on the specifics of the gossip protocol. I don't know how often they happen in practice.
I'm a bitcoin newb, so this might not actually be a weakness -- if it's not, I'm interested in knowing what I don't understand!