The brown layer of oxide has an almost-negligible and certainly unquantifiable (for water coolers, anyhow) effect upon performance. Do not worry about that - it'll form quickly no matter what you do.
Well, pretend you have no cold plate. This is a tradeoff, because you lose the thermal interface material (TIM) joint between the processor and the TEC (two TIM's, in fact) BUT the heat from the processor will be applied to the TEC in a 14mm*7mm (how big are processors these days?) rectangle, only engaging a fraction of the thermocouples in the TEC. If we apply a very thin (basically, of zero height) coldplate, you will have two TIM joints, but the heat will still be applied to the TEC in a 7*14mm rectangle, engaging the same fraction of the thermocouples, since the heat has no chance to spread at all in a cold plate of zero height, and thus it will be worse than no cold plate at all. Since heat spreads in a conic fashion, with the CPU as the tip, as the cold plate gets thicker, the rectangle of thermocouples engaged will grow and more and more will be engaged, which intuitively implies that cooling efficiency will increase. Copper has practically no thermal resistance internally, so that won't impede cooling, and time has no effect, since we are running steady state systems.
There is no "best" thickness for a cold plate, you'd have to make the decision yourself.
I'm guessing that the above will do nothing but confuse you, but I'm trying.
http://www.overclockers.com/articles305/ has information on the matter - lower thermal resistance is better. The thickness of cold plates has diminishing returns.