Communication is a physical process. Unless you can put forward a coherent, testable model for non-physical communication, then talking about communication from a non-physical entity to a physical entity has no semantic meaning. If no experiment can be performed to distinguish two hypotheses (e.g. that there is or is not such a thing as an epiphenomenon) then that thing is irrelevant given that human minds are purely physical objects and human thought, as far as all evidence is concerned, obeys our best models of computation (Church-Turing thesis, etc.).

Epiphenomenal hypotheses are still required to pass Occam's razor. If there is a simpler explanation (e.g. purely physical) that accounts for the evidence, then intellectual integrity demands you take that view. Positing epiphenomena is no different than positing unicorns, unless you have quantifiable evidence for the phenomena and hence they would not be epiphenomenal.