Did more work on the method. Well, as it seems, for 2-3 dimensional problems you probably won't find a better method. I have to work to make the method converge faster on higher-dimensional problems. But it's quite good for <=10 dimensional problems.
* * *
This strategy was tested on 300+ classic 2-10 dimensional optimization problems and performed well. Due to its design this strategy may be particularly good at improving an existing sub-optimal local solution. This strategy offers a very fast convergence on 2-3 dimensional problems, moderate speed of convergence on 4-10 dimensional problems, and very slow convergence speed on >10 dimensional problems, usually 10 times slower than the best competing strategies. However, on 2-3 dimensional problems there is little competition to this strategy.
This strategy was compared with results of this paper (on 242 published non-convex problems): http://archimedes.cheme.cmu.edu/?q=dfocomp
This strategy was able to solve 63% of problems in 10 attempts, 2500 iterations each. For 2 dimensional problems, this strategy's success rate is 96%. For 3-9 dimensional problems the success rate is 60%. In overall, these results place the strategy on 6th place among 23 different strategies. On more that 9 dimensions the results of the strategy are currently quite poor: this strategy requires much more than 2500 iterations to converge on the solution.