The above mentioned tennis-player would be much better off if his perceptual systems were smart enough to make use of this geometrical invariant. (Runeson, 1977, p. 176).I think that is not such a promising line of argument for a smart perceptual mechanism. The problem is that even if we grant that an agent would be much better off performing a given task were she to be able to use a geometrical invariant, there are other considerations that might prevent the "smart" mechanism from being adopted. For one thing, solving this particular task may not have been evolutionarily significant. For another thing, there could be other selection pressures in play. For a third thing, there could be physical or biological constraints that inhibit the adoption of the smart mechanism. (To comment a bit on this last point, it might be smarter for human babies not to be born so immature, but being born immature seems to be a constraint imposed by having to get the baby's head out of the human birth canal before it gets too big to get out at all.) And the fact that we have relatively little knowledge of what those constraints might be encourages the view that there are no such constraints.
So, I think that this line of thinking is, as I've said before, a little dicey. Maybe it would be better to have someone with more experience with evolutionary biology or philosophy of biology comment on this. Old fashioned experimental work seems to me more reliable here.
Runeson, S. (1977). "On the Possibility of 'Smart' Mechanisms" Scandinavian Journal of Psychology, 18, 172-9.