The empty return trip would not generate any income, so unless the cost of shipping the cargo is greater than the income, having the vehicles loaded in both directions would still be beneficial. If it is, sending the vehicle empty both ways would be the best (except not having the vehicle at all).
Sure. But currently, the empty return trip costs just as much as the full trip. Hence if you could fill the vehicle on the way back, you can double the income with no extra costs. If an empty vehicle had only half the running cost (to keep things simple), the empty way back would still cost money. If you fill the vehicle on the way back, you still double your income, but you also add 1/3 in costs, which is beneficial, just not as much.
Usually, to double-fill a vehicle, you need to utilize hubs, which means taking a detour. In some settings, a detour does not get paid. So using hubs is a balance between the benefit of double-filling vehicles and the drawback of having a detour. If the benefits are reduced, the balance shifts, and hubs become a bit less profitable in comparison to direct connections, hence there will be less situations where they are the right choice.
Simulation of fuel cost probably would not affect train length. Nor do I think that is a major influence in real life, beyond the cost for the fuel used for moving the locomotive itself having to be covered by the profit from the following wagons, which is kind of already in Simutrans.
It does affect train length if you model it to affect train length. To make it simple, let's say a locomotive has a base fuel consumption value, and each wagon multiplies to that value. Let's use a rather high number for larger affect - we multiply by two.
If the locomotive alone costs X for fuel, the same locomotive with 5 wagons would cost 32X for fuel, about 6X per wagon. With 6 wagons, it would cost 64X fuel, about 10X per wagon. If a full wagon would generate 8X income, you could have a train with five wagons and make a profit, while a train with six wagons could never do that.
The difference is huge if you double the fuel cost. Naturally, the multiplier would be more like 1.4 - 5 wagons would cost 5,38X fuel, 1.08X per wagon; 6 wagons 7.53X, 1,26X per wagon; and 7 wagons 10.54X, 1.5X per wagon. So with each wagon you add, you have to pay more per wagon, earning less per wagon. But on the other end, you have the existing mechanic of paying the locomotive with income from the wagons which becomes less per wagon the more wagons you have.
So you don't want to have too many wagons and you don't want to have too little wagons, there is an optimum of wagons somewhere in the middle. And the more you diverge in either direction, the less profitable your convoi will be. This is different from the current concept, where you want to have as many wagons as possible until you hit the rather harsh limit of getting slower. Sometimes it might even be beneficial to have a convoi that can't even reach max speed anymore.
If the goal was to shorten trains, you might as well change the power of locomotives. But that would be a hard limit. With such a fuel concept, it would be a soft limit - which are arguably better, as they provide model railway players more leeway in doing something besides the intended (for a cost) and 'power players' more options to optimize.
Whether it has anything to do with how fuel works in reality I can't say (I know it would work somewhat like that for speed - the faster, the more fuel you need. Is it the same for weight?)