At the moment there doesn't seem to be any specific suggestion in the RFC for implementing this.
At this time the only things I can see that contribute to the falling off of earnings is that small vaults lose out on the opportunity to serve data because of their small size yet they still contribute to all the caching, routing, consensus etc. For the "oversized" vault I am not sure, until it reaches a point where it is serving up so much data that the bandwidth becomes an issue. (this is true for 0ne large or equivalent in overall size multiple vaults)
for the oversized vault compared to an average vault, they will fill at similar rates so then their earning ability is similar till the average vault is "full" after which the oversized vault then has more chunks thus more earning capability. How long does it take to fill an average sized vault though? So the disadvantage to the oversized vault is that they waste so much potential earning, they would be better off running a number of average vaults and thus contributing more to the network (multiple nodes -caching, consensus, routing, etc)
At this time (& current RFC) the reason Joe is better off is that Joe isn't being charged anything to run the average vault because it is spare resources. But Dave using the Dwarf's data centre is also using spare resources but is 3 billion years away in space. Now Dave the data centre owner has to pull resources away from earning $$$s elsewhere to use for the SAFEnetwork and also has to pay the aircon, electricity, staff, building rental and all the other expenses. It might not even be profitable considering the rate of coin earnings and expenses. Let alone the earnings he loses because he pulled the resources away from earning real $$$s from people using the datacentre