Erratum:

As observed by Frits Vaandrager about a month after publication, my new term model is not isomorphic to the one of [TCS51]. For divergence-free process the models are isomorphic, but in general my model makes strictly less identifications. In particular, contrary to what I claimed, my model does not satisfy KFAR, whereas the one from [TCS51] does. It also does not satisfy the weakenings of KFAR discussed in the paper, but it does satisfy the weakening KFAR- as proposed in the cited paper by Bergstra, Klop & Olderog (1986).
The results concerning the relative inconsistency of four widely used proof-principles are unaffected by this mistake. The new restriction of AIP is valid in both the new model and the model of [TCS51].
The easiest repair would be to drop the new model from the paper and replace it by recalling the model of Baeten, Bergstra & Klop (1985) [TCS51]. That model is based on Milner's notion of weak bisimulation semantics, and additionally distinguishes deadlock from successful termination.
However, the new model is interesting in its own right, even when it can not be seen as an alternative presentation of the one from [TCS51]. In particular, Bol & Groote (1990) [The meaning of negative premises in transition system specifications, ICALP'91] define a compositional priority operator in this model that allows silent actions to have priority over certain visible actions, which would not be possible in weak bisimulation semantics as employed in Milner (1980) or [TCS51].