Experience has borne out that members of the first machine class are the ones which we should consider reasonable models of computation in the course of formulating the Cobham-Edmonds Thesis. It is also widely believed that members of the second machine class do not provide realistic representations of the complexity costs involved in concretely embodied computation (Chazelle and Monier (1983), Schorr (1983), Vitányi (1986)). Demonstrating this formally would, however, require proving separation