The retail FT3 is limited by the basic 32-bit Windows subsystem to 2GB of total working space. You can set the /LARGEADDRESSAWARE flag (Google for what and how) on your FT executable to use 4GB on 64-bit Windows. The 32-bit version of FT3 that I'm using already has the /LARGEADDRESSAWARE flag set and I forget that sometimes.
For your dropoff problem, make sure that you're blurring the channel that's causing it. Your dropoff could be due to changes in the offset channel, prescale offset channel, or even roughness channel; blurring a channel without data in it will have no effect, regardless of how much you blur it.
Neither Wilbur nor FT have a feature of the sort that you describe. It should be possible to implement such a thing, but it would probably be a bit clumsy to operate. The previous tutorial about how to make a flat-topped mound is an example of a set of workarounds to get that sort of effect with the current implementation. You might also be able to use the profiled mound option with most of the second part being a flat section. Still clumsy to operate, though.
When you said that you were having problems with add to selection and subtract from selection a few pages back, I thought you were referring to Wilbur. I may have forgotten to hook up that setting for that particular feature in FT. It certainly wouldn't be the first time that I've made that sort of mistake.
If I recall correctly, there is a recursion limit on the basin filling code that causes it to stop processing at a certain depth (the recursion depth needed is a function of surface size and complexity). You're using a surface larger than about twice what I would normally use, so you may well have hit that limit. For very large surface, I tend to work at a much smaller resolution (1024 or less) and then scale up the data to the final form after I get the basic shapes and patterns that I'm looking for.