I have avoided this discussion for a number of reasons, primarily because I am very dubious about the contributions of transaction theory to instructional design. The theory appears coherent, and it certainly represents a conceptual elaboration of component display theory. The idea of transactions representing different kinds of relationships is great. Operationally, though, I don't see any advancement beyond Component Display Theory (CDT). I have never seen a full set of lessons produced using transactions and certainly no research validating its efficacy. In fact, there is precious little research on CDT (half dozen dissertations, I think). I may simply be ignorant here (I do that well), so if there is research, I'd be interested in reading it.
My reservations, however, result from attempts to implement it. My research assistant at Denver, Woody Wang (now at George Mason, at least for the moment) and I, developed a Knowledge Analysis System in HyperCard based on transaction theory. The idea was an instructional transaction generator. In trying to produce transactions, we discovered that students were very confused about them. The only examples that were available were for those transactions that corresponded to remember and use concepts, procedures, and rules (i.e., CDT). Students were able to produce some instructional prototypes, but what we all learned was how fragile the theory was (as were our HyperCard stacks). We shifted gears, and developed a set of SemNet (semantic networking program from San Diego) shells with pre-defined links representing the necessary (we thought) propositions that should form the bases of transaction shells. Students populated these shells with content (we have some good examples of these) and then converted the SemNets to HyperCard prototypes using a neat HyperCard program called Wayfinder that was developed by Brock Allen and his RA. So we actually have instruction (so to speak) developed using transaction theory.
However, our success in generating prototypes posed another vexing question, namely the integration of transactions into meaningful instruction. That is, what combinations and sequences of transactions comprise what kind of instruction. Learner control is a convenient answer, but we felt that certain transactions probably shouldn't be linked to other transactions. We found no help here. So, we did what any red-blooded American would do, we gave up. The semester was over, and it was time to go skiing. No seriously, we concluded that transaction theory needed more development before it was a usable theory (i.e., one that could develop reliable, coherent instruction, which is the putative advantage of objectivist designs), and also we were devoting more of our energies to designing constructivist environments. It was an interesting exercise, and we learned some things about design. And it may be possible for Dave and company to develop transaction theory well enough to implement an automated ID system, which I presume was its goal. For all I know, they may have already done that.
Instructional design models, especially objectivist ones, need to be fairly bullet-proof in order to produce good, reliable instruction. As of a couple of years ago, transaction theory, to our reckoning, had not arrived at that point.