It's possible to model the changes in allele frequency as a diffusion process. Diffusion processes have things happening at all "distance" scales.
A randomly generated change in allele frequency in conjunction with selection doesn't yield a "random" result as you point out. I was surprised at the speed of "adaptation" though. Originally I expected "drift" to be proportionaly to Sqrt(Time) and selection to be proportional to Time. However, the change in allele frequency (measured from an hypothetical ideal frequency to fit selection function) can be exponentially fast.
Cool.
Do the models fit real-life data, especialy regarding phenotype?