I've not seen one of these failures in the flesh, only in pictures. But it seems to me that they seem to have both breaks in the aluminium and areas where the adhesive bond has failed. It isn't clear to me that one always precedes the other, but there are two hypotheses which might explain most of the breakages;
1) that the adhesive bond begins to fail (e.g. at the spider end) and once it gets so fair up the crank the back part of the crank cracks.
2) that the back part of the crank cracks first and this then causes the bond line failures in the adhesive joint.
Quite a few of the failures appear to show corrosion within the failed adhesive bond, which suggests that this may have initiated the failure.
However that the back of the crank cracks isn't any great surprise to me; the back of the crank appears to have sharp edges to it (presumably so that the bond line doesn't show) and even the slightest roughness along these edges will be a perfect crack starter.
There are a few failures where the outer part of the crank cracks near the spider (a traditional weak spot in some cranks) and this combined with the corrosion in the bond line makes me think that this is where the failures most often initiate, probably first in the bond line.
So why do these failures occur at all? It could be one of three likely suspects
1) marginal design
2) inconsistent manufacturing
3) service conditions
or a combination of the above. One of the things that rings warning bells with me in relation to the first two is that the joint profile is complex and the failures appear to show that the adhesive layer isn't of perfectly uniform thickness; in fact it probably can't realistically be, simply because the profile of the parts is so convoluted. The adhesive in such joints is by far the least stiff part of the assembly, which means that it strains more under the applied loads. Anywhere the bond line is very thin, the strains in the adhesive layer are highest and this is where the bond will fail first, all things being equal. Adhesive bonding systems vary considerably but there are all kinds of tricks involving fillers of defined partical size and spacers (of matching stiffness) that sit within the bond line which are claimed to help.
FWIW it might seem like a good idea to add bolts, rivets, etc to such joints as 'belt and braces' . This can indeed give a more benign failure mode (the parts might be made in such a way as when the bond fails the crank flexes obviously before it comes apart) but such designs are actively resisted by adhesive bonding specialists; they argue that the parts see such high stress concentrations at the fasteners that failures are made considerably more likely by the presence of these additional fasteners etc, and that the resultant joint is much weaker than even one with very variable bond line thickness. They are probably right, too, but if the failure mode in a plain adhesive bond might then kill folk maybe it isn't a good trade after all....
If cranks never ever break then they are built too heavy....(I can't think of a racing type crank design where I have never seen a failure) but it is difficult to know if the failure rate in this design is acceptable or not.
cheers