In two-way shape memory alloy, it can memorize not only shape of Austenite phase, but also shape of Martensite phase.
Then, if NiTi has R-phase, can it memorize shape of R-phase likes triple-shape memory effect?
But if you do have regular 2-way SME and have the R-Phase as he sais, which should have a defined shape in between the austenite and martensite shapes. So I guess under the right circumstances (exact temperatures, material training, etc.) you could have 3 predestined shapes, but I can not see how this could possibly be used, because there would only be a usable force transforming to austenite. Also I'm not really sure if a pure R-phase can be produced reliably at all. And since it would always be an in between shape you could get there easier by controlled partial activation.
Apologies for my previous comment which assumes third stage not as a transient, intermediate state but a phase domain with its definiteand distinct properties For your definition of third stage, see the following for an example of triple-shape memory effect.
Materials Today, 2010, vol.13 Number 7-8, pp.54-61