Our second series, “Rfi” (intra-thread, or internal, read-from), exercises the impact of atomic variations on store forwarding. To do so, we generated a restricted subset of our plain tests, namely some that exhibit a violation of SC in the form of a critical cycle, with at least one internal (intra-thread) read-from edge, and up to 3 threads; this gave a series of 1338 tests. We replaced the accesses along the internal read-from edges with RMWs and load-reserves, as in the first series. Interestingly, many of the Rfi variations that involve only load-reserve (as opposed to RMWs) were observed on Power 6 yet not on Power G5 or 7. This suggests that the implementation of load-reserve is less liberal on the latter two. We provide experimental data that support this suggestion in a dedicated section.