I'll try to stay tight to your "science" and "evidence" clauses, because there's a whole lot of non-science reasons to use free weights. (Just off the top of my head, there's cost, versatility, compactness, and portability.)
This roundtable discussion (PDF), with copious references, is one of my favorite sources for the machines/free-weights science. Among the points made therein:
- (Wathen) "There is some evidence that more carryover is likely with free weights than machines when training and testing in both modes." (page 21)
- (Carpinelli) "...there were no significant differences (P < 0.05) between the [free-weights-trained or machine-trained] groups for any of the [strength] variables. It appeared that vertical jumping ability was improved to a greater extent by training with free-weights and the Universal apparatus than by training with the Nautilus device. However, the differences are relatively small, and in view of the similarity of the strength gains, the practical application of the finding is unclear” (ibid).
- (Carpinelli) "The absence of any supporting evidence suggests that when used properly, free weights and machines produce similar results in a healthy, athletic population." (ibid) While that's true, I would argue that the very act of choosing machines produces a radically different and markedly inferior program due to choice of exercises, but I can't prove it.
- (Wathen) "Augustsson et al. (3) indicated that free-weight squat training has more impact on vertical jump performance than does a program of isokinetic knee extension and hip adduction. There are few studies in this area, and most are equivocal, but more evidence points to specificity of muscle action type and movement pattern favoring dynamic freeweight movements that closely approximate the sport movement (5,7, 15). After all, most sports performance is dynamic, multijoint, multiplanar, requires balance, and deals with both eccentric and concentric actions (3, 10). Sounds like free weights." This is a great passage.
- (Stone) The same conclusion, stated more succinctly: "The major contributing
factor to the superiority of free weights compared with machines is the ability of free weights to mimic and overload most athletic (and daily task) movements. Because of this aspect, there can be a greater transfer of training effect."
All in all, it is a balanced discussion with tons of studies to curl up and read.
As to your specific questions, I do not see how the fact that most middle-class Westerners sit a lot and are weak has anything to do with how we evolved. I would venture to say that we know with fair certainty that our ancestors mostly got strong by lifting things, carrying things, and running. I fail to see the relevance of this point to whether machines are empirically superior to free weights, except to point you to the inherent joy of training and playing outside. (Erwan le Corre has more to say on this topic.)
To address your point about cables and machines being different enough to produce different stresses: most machines, though less so with cables, dramatically reduce the number of muscles involved in a motion. They are designed to isolate muscles instead of allow the body the act as a whole. So, yes, the leg curl machine might provide a different stressor to more strongly incite growth in the hamstring, but as noted in the references above, that stressor is highly specific and has less carryover to everyday life and sport.