A few comments on the choice of simplex method

If the objective is to improve the yield and to locate a near-optimum experimental domain, any of the simplex methods will do the job. 

If the objective is to establish the maximum possible yield, none of the methods will achieve this without an excessive number of experiments. Response surface modelling is often more efficient for this purpose. 

If no such details as to the effects of the experimental variables are known, it is a matter of taste which design should be used to define a starting simplex personally I prefer the simplex design given above. 

---