My understanding is that when two pools find the 'same' block, they are actually finding different hashes that meet the difficulty requirement (please correct me if I am wrong). In this case, given that the network difficulty makes this so that the expected block length is ten minutes, and invalid blocks are not that common (IIRC, we have seen two in the last month or so), I am quite surprised that there should be such a 'collision' on a two-second block - surely the odds of this happening would be vanishingly small, unless there is something intrinsic that makes the difficulty on certain blocks much lower than it should be.
Not only different hashes; each pool assembles its own block from the available transactions in the transaction pool according to its rules. So the "same" block can potentially differ between the two pools in the sense it contains different transactions.
The last idea is quite interesting. Can someone have enough will and power to check it? :-)
[edit] I'd sure not say
vanishingly, but the chance is of course smaller for two pools at once than it is for one pool.