I agree with Jeffrey Beall’s reply. In addition, I believe that there are already so many journals (at least in my area of expertise) that one should have very good reasons to create a new one.
A good example of a success story, in which a journal managed to establish itself by word-of-mouth advertisement, is
the case of Markov Processes and Related Fields. See Vadim Malyshev’e editorial in a recent anniversary issue at
http://mech.math.msu.su/~malyshev/abs10.htm
