I shall use the phrase "non-derivatively libertarian-free" (NDLF) to describe a libertarian-free choice that does not inherit its freedom from earlier free actions. This corresponds to Kane's Self-Forming Actions. Now consider this plausible principle:
Thesis 1: If x NDLF-ly chooses A in circumstances C, and p is a proposition explanatorily prior to x's choosing A, then were x not to have NDLF-ly chosen A in C, p would still have been true.
A consequence of this is the following PAP:
Thesis 2: If x NDLF-ly chooses A in C, then x's failing to NDLF-ly choose A in C is logically compatible with any proposition that is explanatorily prior to x's NDLF-ly choosing A in C.
(The argument from Thesis 1 to Thesis 2 is this. Suppose Thesis 2 is false. Then we have a proposition p explanatorily prior to x's NDLF-ly choosing A in C such that p entails x's NDLF-ly choosing A in C. But then x's failing to NDLF-ly choose A in C entails ~p. It is obvious that if x NDLF-ly chooses A in C, then x's NDLF-ly choosing A in C is not logically necessary. But if u entails v, then at least if u is contingent, were u to hold, v would hold. So, were x to fail to NDLF-ly choose A in C, then ~p would hold. But by Thesis 1, it follows that were x to fail to NDLF-ly choose A in C, then p would. But these two conditionals cannot both be true if the antecedent is possible, as it is. So Thesis 2 cannot be false.)
Now on to the argument. If Molinism holds, then the following scenario is possible:
Scenario 1: God believes that were he to place agent x in circumstances C, the agent would NDLF-ly choose A in C, and for that reason God in fact places agent x in circumstances C.
Now, assume that if p and q are explanatorily prior to r, so is the conjunction p&q. Suppose Scenario 1 holds. Let p be the proposition that x is in C, and let q be the proposition that God believes that were God to place x in C, x would NDLF-ly choose A in C. Then p and q are explanatorily prior to x NDLF-ly choosing A in C. Hence so is their conjunction. Hence, their conjunction does not entail x's NDLF-ly choosing A in C (by Thesis 2). But, necessarily, God believes only truths. So, q entails that were God to place x in C, x would NDLF-ly choose A in C. By modus ponens, p&q entails that x NDLF-ly chooses A in C. Hence, p&q both does and does not entail that x NDLF-ly chooses A in C, which is a contradiction.
This is, of course, a version of Adams' circularity-in-the-order-of-explanation argument. Strictly speaking, it doesn't show that God can't know conditionals of free will, but only that it is incoherent to suppose him to act on that knowledge in the way indicated in Scenario 1. Thus, the argument is compatible with a weak Molinism on which God knows the conditionals but must bracket that knowledge when choosing to act.
I actually don't quite buy the argument because my current view of counterfactuals does not support Thesis 1 (but neither does it support Molinism).