Hi Sanne,
There is no way to get a cluster-level p-value at the moment. If that's what you absolutely need, it's simply not possible to do directly or with simple steps. Surely you could trick randomise, save all possible permutations, convert the stat to a p-value, take the minimum of each, then convert to a z-score, compute the cluster sizes for each permutation, build a null distribution of the largest cluster size, then confront the result without permutation to that null, etc... Not only this will require a lot of work, but it also requires that the set of permutations needed to test both contrasts overlap with each other. This is trivial if they are both continuous regressors, but if both are discrete, and perhaps orthogonal, even what I just described becomes impossible.
So... no cluster-level p-values for the intersection of contrasts (by intersection I mean binarise both and multiply one by another, i.e., the overlap).