The real issue, I think, is not the fact that mail is counted in "bundles", but the fact that there is so much less mail generated overall than passengers. Mail behaves differently to passengers in Experimental in that it does not refuse to be sent if it takes too long to get to its destination, although it will, like all cargo, take the quickest route. Mail thus behaves more like passengers do in Standard than like passengers do in Experimental, and is subject to a much stronger network effect, where the amount of mail carried increases exponentially as the number of buildings producing and consuming mail increases linearly. This means that, just as in reality, long distance mail transport potentially needs to have a very high capacity on highly connected networks, whereas local mail transport and mail transport generally on early/lightly connected networks need not have anything like that capacity. This, again, reflects real life. The large capacities for mail that one sees are based on the potential need to carry many bags (10x bundles) of the stuff in the case of a trunk connexion on a highly connected network. These are, in so far as it was possible to research them, the real life capacities of the vehicles. If the capacities of mail vehicles were reduced tenfold (or the amount of mail generated increased tenfold), the ability to carry very large quantities of mail on trunk routes would be lost, unrealistically.
The problem with setting percentages of capacity is different. As Vladki suggests, in theory, that can be solved by allowing different percentages for each type of load, although actually implementing that is potentially complex, and there is an extraordinarily long queue of higher priority tasks.