Perhaps a controlled random source would work. Something that operates along the lines of the Buchla "Source of Uncertainty", such as Doepfer's A-149 modules, Mutable's Marbles, or the Frap Sapel. By using something like this, it's possible to constrain and direct the random variance into a specific "zone" of random activity. Within that zone, you'd still have varying degrees of randomness, but with the additional control functions, you get a more workable stochastic result.