I think Feynman had a go at this one, can't remember exactly.
Say we've got our 2 photons, seperated by a galaxy. Our friend in the next galaxy is waiting for our message. Say we agree, before we seperate, that up means "yes" and down means "no". Our friend wants to know if George Bush won the election.
The moment the election results come out, we measure our photon, setting the spin of the other photon. Now, the measurement that we make is random - we don't know whether our photon will be up or down. Therefore we can't send a message with this technique, because we cannot determine what our friend will see when they measure their photon, because our spin (and therefore their's) is random.
Likewise, we cannot say to our friend "You measure yours at time x and if it's set then Bush won, otherwise Kerry won" because the act of measurement collapses the relationship and causes OUR photon to be set! The friend could not tell whether we had set ours or not.
The whole problem here is that the act of measurement can only be done once, and it is impossible to tell when that measurement has been done, and impossible to control what that measurement will be*. In short, we get information about the particle that is light years away, but our friend cannot get the information from us faster than the speed of light. We can't use another faster-than-light entanglement because we get the same problem - to transmit meaning of the result, we must transmit information, which can only be done in a non-random manner, and that means controlled interaction, which excludes entanglement (which is random).
The whole point is that although we can gain information about something far far away, we cannot send that information far far away faster than light.
* Although we cannot control whether our photon is up or down, it is possible to 'turn' a photon. If we have two polarizers at 90 degrees to each other, no light can get through. It would make sense then, that if we slotted another polarizer in between these two, at 45 degrees, then still no light would get through. But in fact, some light DOES get through then. It's as if the light is being turned by the filters. This happens not only with regular light, but also with single photons. Now as long as we don't measure whether the light gets through the filter array, we don't collapse entanglement, hence we should still be affecting the far away photon. I don't think this helps though, because as soon as you try to measure the photon's polarization, you collapse entanglement and the above paradox of being unable to transmit information remains. It's too late for me though, can't think properly, so any thoughts anyone?