In fact it is extremely simple and obvious.
After the user has made two clicks defining the first fillet the location of where the user must click next is completely known.
A rolling ball fillet always ends at a surface edge. If there is another surface at this edge point that has the same surface normals that is all the information that is needed for Rhino to make the next filletsrf in the sequence. The same thing for the next fillet and the next all the way around.
There is no good reason why the user has to tell Rhino where to make each fillet. Are you saying it is difficult for the developers to figure out where a corner of a fillet is located? Are you saying that it is difficult to determine if there is another surface at that point that has the same surface normal as the corner of the fillet? What part of this are you claiming is difficult?
Rhino can make rolling ball fillets as good or better than any solid modeller. The problem is that to get those excellent fillets the user has to do all the tedious repetitive work that should be done by the computer. And that is only because no development time has ever been devoted to automating those repetitive tasks.
That statement is completely wrong. It is much easier to do this on a surface to surface basis. The only reason there has been no progress in filleting in Rhino is exactly because there has been zero effort put into doing the very simple and obvious things that would make Filletsrf a very powerful tool.