All this back and forth...
Seems simple to me. Pack as much speed and memory as physically possible within a given price. RAM is pretty cheap these days. If, for instance, you can add 2GiB of memory and an 060 class CPU for even twice the price of a 128MiB based solution then do it. In reality I doubt it will really increase the cost by such a large factor.
2GiB may seem vast for a 68K machine but it simply means applications can work on larger projects and more applications can be open at once (as long as they are not CPU bound).
One technical point, if you are implementing an MMU, you might get some big tables.
If there are other, hardware based reasons for sticking to smaller memory sizes, then that's fair enough.