Here is my take on the question.
An early theorem in propositional logic (or digital logic, or gate based circuit design) is that
any Boolean function has a representation in disjunctive normal form. A somewhat
cumbersome example is XOR(x,y,z) = xyz + xy'z' + x'y'z + x'yz', where I use + for an OR gate
and concatenation (instead of explicit multiplication $\cdot$) for AND and ' for NOT. If I
have a collection of a large (but fixed) number of NOT gates, followed by an array of four
AND gates, each with six inputs, and an OR gate with four inputs, I can wire those together
in a configuration that, once I assign signals to the inputs appropriately, allows me to
represent any of the 256 three input Boolean functions. This configuration is a gate based
realization of a Universal Boolean function for n=3.
A bit of personal history: I was involved in a project decades ago which allowed digital logic
designers to use software to help design and simulate circuits created with PLDs. These
Programmable Logic Devices were modifiable gate arrays, and often served as
physical realizations of approximations to Universal Boolean Functions. I have not seen the
Preparata and Muller paper above, but I would bet a small amount of cash that the paper
inspired the use of gate arrays in chip and circuit design in the folllowing decades.
Gerhard "Practical Uses Of Clone Theory" Paseman, 2013.05.25