I agree with Tattoo that goods requirement could be higher and this would make our train networks look much prettier. However I wouldn’t want production to ever exceed demand.
At the moment I prefer running trains the full length of the production cycle. My favourite routes go from raw materials, around an industrial zone to a production facility, then drive back with the goods into the industrial zone on the way back, before returning to the raw goods spot. I use mixed trains (everything except wood/goods combi needs more types of carts), Even though every part of the trip half of my carts are empty, only the final lap (usually aprox. 25% of the line) all my carts are empty. (50% raw -> production, ~15-35% production > delivery, what’s left is completely empty.) Because a set of carts is cheaper in maintenance than a whole new engine, this is in my experience an effective way of reducing the time that all carts are empty. Though this is a tactic that only works when demand is sufficient, I find it to be very profitable and it is so far the only instance where I run trains with more than one cart type. (I find that visually pleasing).
An early train cargo tip is to use signals to space your trains. Literally disect it in the same number of sections as you plan to run trains on it. This optimises spacing, which makes the frequency time more accurate and allows you to run as few trains as possible, with as much carts as needed, to allow for optimal profit.