Is wavefunction collapse “global”? I have the feeling that the premises of this question may not be coherent (so to speak), but here goes:
Suppose we have a system $X$ in a quantum superposition between states $0$ and $1$, say, with equal amplitudes. Suppose we have two observers, Alice and Bob, who have never interacted with each other in any way. Alice measures $X$ to be $0$. Then, later, Bob measures $X$.
Will Bob “always” measure $X = 0$ if he never interacts with Alice, or will the wavefunction not have collapsed for him yet? Or does their both having measured the same system mean that they've interacted in a way that ensures they'll observe the same thing?
 A: Wave function collapse is not global, it is fictional. Let's suppose that the state is $\alpha|X=0\rangle+\beta|X=1\rangle$, where $|\alpha|^2+|\beta|^2=1$. 
When Alice measures the state, an operation is applied that correlates both Alice and the environment with the value of $X$, like so $|X=j\rangle|0\rangle_A|0\rangle_E|0\rangle_B\to|X=j\rangle|j\rangle_A|j\rangle_E|0\rangle_B$. The environment is just everything around the system other than Alice. The $A,E,B$ subscripts stand for Alice, the environment and Bob respectively.
Bob might get the measurement result directly from the system, or from Alice or the environment. In any case, the end result will be
$$\alpha|X=0\rangle|0\rangle_A|0\rangle_E|0\rangle_B+\beta|X=1\rangle|1\rangle_A|1\rangle_E|1\rangle_B$$.
After the measurement there are two versions of Bob: one version sees 0, the other sees 1. There is no version of Bob that sees both outcomes or some weird mix of 0 and 1, and there is a large literature that explains why this is the case, for an example see
http://arxiv.org/abs/quant-ph/0703160.
The short version is that only information contained in the eigenvalues of some observable, or a subset of such information, can be copied from one system to another. Bob won't see any other information because none of the rest of the information will be copied to him. This follows from quantum mechanics with no collapse postulate. So the collapse postulate is unnecessary for explaining that result.
A: Here's how I understand your question: 
A and B are space-like separated and make a measurement on a single particle that has equal (or just non-vanishing) probabilities of being in A's or B's region.
You now ponder how the measurement process works on a deeper level.
Could the collapse be a dynamical (i.e. time dependent) process? I think it can not. If it were, A and B would both try to 'pull' the wave-function to their side. You also get in trouble with relativity and the question who starts the process first.Thinking deeper, this will probably lead to the conclusion that the wave-function is not to be considered as 'real' (or 'ontic'). But I have not done this or seen this done.
It is thus better to regard the wave-function just as information. As soon as A or B have measured the particle, they have gained information and can update (=collapse) the wave-function.
Back to the specific question: For the physical outcome, it does not matter WHEN you collapse the wave-function. QM always assures that only one of A,B can measure the particle. Asking 'when' the collapse 'occurs' is not a sensible question, because the collapse is not dynamic.
Added: If B knows the outcome of A, he must use the 'collapsed' wave function. If he does not, he must use what he knows, i.e. the original state X.
