Extremum-seeking methods are unconstrained real-time optimization techniques that control of the gradient to zero. The crucial difference between them lies in the gradient estimation method used. Multi-unit optimization technique proposes the use of a multiple units operated with an offset between them and the estimation of the gradient is by finite difference. Though this method gives fast convergence, the major bottleneck is that it assumes the units to be identical. This paper addresses the case where the static curves are indeed identical, while the dynamics are not so. It is shown that if all the units are stable, despite the difference in dynamics, the method would indeed converge to the true optimum. Also, it is shown that the difference in dynamics does not affect stability in the neighborhood of the optimum. In addition, this paper presents a possibility of replacing real units by static models in the calculation of the gradient. Experimental results are presented from a mixing system where an optimal temperature is sought.