Incidentally, the success rate of fundamental research for a given finite time horizon k is an untestable quantity.

How exactly do you measure "success" in this case ? As for me, I find myself hard-pressed to think of any examples of fundamental research that weren't ultimately beneficial -- except perhaps for instances of outright fraud or gross incompetence.

Even if a scientist spent five years and a million dollars trying to discover, say, the link between gene X and phenotype Y, and found no such link, then the work was still not in vain. Firstly, we can now be more certain that gene X does not cause Y; secondly, we can most likely gain a lot of collateral benefits from the work, leading to an increased rate of discovery in the future.