Your argument contains a fallacy. You are assuming that mental agency, if true, would operate indiscriminately –or at the least, indiscriminately within the confines of the body. In other words, if I concentrate, I should be able to dissolve my hand into its component atoms as easily as I can raise it or lower it, and no one should ever suffer from paralysis, etcetera.
This does not necessarily follow from the idea that mental agency might exist. For example, I can construct a scenario that would account for the given data without leading to that situation.
In general, events on a subatomic level do not create visible effects on the world, because subatomic particles are so tiny an numerous that their effects never aggregate to the point where it becomes a macroscopic causal agent. Even an occurrence as momentous as the decay of a radioactive atom will generally pass unnoticed by the world.
Yet it is possible to create circumstances under which even that single atom’s fate can have an outsized impact. For instance, the decay of a single atom in an armed nuclear bomb creates an explosive chain reaction. Similarly, a Geiger counter, which is designed to measure the decay of single atoms, can be attached to an apparatus that responds to the atom’s decay in a macroscopic way –by killing a cat, in Schr0dinger’s famous example, or less violently, by playing a recorded song, or raising a mechanical arm.
We would call the decay of the atom the causal agent of the death of the cat or the raising of the arm, but that would not imply that the range of agency of the atom was unlimited. The decay of the atom could only create a tightly defined range of macroscopic effects predetermined by the nature of the apparatus.
In theory, human agency could follow an analogous pattern, with the brain taking the place of the Geiger counter and the body being the apparatus that translates the subatomic event into a macroscopic action. In your version of Epiphenomenalism, you admit mental processes, so let us further theorize that a mental process somehow localized in the brain could create subatomic events of the type the brain was optimized to detect. The end result would be that the mental process would indeed have macroscopic agency, but only though the well-defined channels created by the combined apparatus of the body and the brain.
I’m not claiming that this is in fact the mechanism at work, but it demonstrates the inadvisability of supporting a belief in Epiphenomenalism through the argument you outlined.