I have to point out that all these folks talking about "see, this is why I don't use carbon" or statements to that effect have a complete misunderstanding of why this part failed.
I'm not going to fully explain the way carbon structures derive their strength, or why, but the information is freely available on the net as well as in numerous books to be found at any bookstore. Do some research on composites before making sweeping statements regarding their suitability for a particular application.
As far as this particular failure, I can make both some very specific and very general statements about it, both of which I expected to have already been made, but since they haven't, I feel that someone needs to make them.
First of all, it is quite possible to construct molded structures out of carbon fiber that are quite strong, and extremely light. Arguably, in the case of spider tabs, it makes more sense to design them as a contiguous part of the structure than to bond a metal tab to the primary carbon structure - the bonding point of said hybrid structure being, to borrow a military term I'm rather fond of, an obvious point failure source. Of course, building the tabs out of carbon only works if you design them and manufacture them correctly.
What is 'incorrect' about the design of the tabs that failed? The first thing that jumps out at me is the edges. Having spent many years working as a tool and die/moldmaking machinist (including building molds for carbon fiber parts), quite possibly the first thing I learned was never to leave a sharp edge, either at a junction between two planes, or around the edge of a hole. In fact, one of my first jobs, just after I graduated from sweeping the floor to actually being able to touch the molds themselves, was to crawl all over the molds with a cordless drill mounted with a chamfer bit, which looks sort of like this: \_/. You use this to put a nice little radius on the edge of each hole. Keep in mind, these were holes no more than .75" in diameter, drilled into billets of high-quality steel that sometimes weighed hundreds of pounds. Still, every machinist knows that any sharp edges will become a focal point for stress, and if you apply enough stress enough times, a crack will begin.
That's my specific commentary about this particular part failure. Now to some more generalized statements, mainly to contradict the aforementioned "this is why carbon isn't good for (insert any bicycle part here)" statements.
Most of my hobbies, and several of my jobs, have been focused on weight. Although I haven't really designed and produced any bicycle parts (modified some existing ones, though), I've done quite a bit of work in the auto racing industry, and gained some experience that I feel is applicable. Despite the fact that our minds are always focused on winning, and in the case of the engineer/fabricator that often means designing or building a lighter part, we try not to lose sight of one fundamental rule: it is better to come in second, or third, or last, than for the driver to impact a wall at 120mph. Thus, while light weight is a primary objective, it can never be obtained at the expense of safety. It is the engineer's responsibility to design a SAFE part. The earlier posts in this thread, one of which I believe said "we are not all about building the lightest possible bike, it must also be safe to ride" and another which sarcastically offered to build a 8/12 spoke wheelset, were both quite correct.
To design a component that can handle the applied loads without failure should be relatively simple these days, at least compared to, say, 30 years ago. Nowadays, computer simulations of stress are extremely accurate - if you have an accurate measure of the maximum load that will be applied to the pedals (say, by Tom Boonen or some other extremely powerful pro) and an accurate computer model of the crankset, pedals, and chainrings (accurate meaning the computer has accurately modeled the shape, thickness, density, and composition of the parts, and can thus determine their resistance to stress), you can do a pretty damn accurate modeling of how they will respond to stress.
Of course, you can't really stop with just a computer model and consider yourself to have a safe product, much as many designers wish it was otherwise. To quote Carrol Smith, who before becoming an author was involved in many winning (and light) auto racing efforts (underlined emphasis is mine; capitals are author's):
All structures are designed so that no component will ever develop a stress or strain that will exceed the elastic limits of the material under the maximum predicted load(s). But, since SOME assumptions are always necessary; SOME joints are inevitable; SOME less-than-optimum section changes; SOME manufacturing tolerances must be allowed; SOME errors will occur in construction/fabrication and post treatment; SOME in-service abuse is inevitable. And, because all intelligent structural engineers are devout cowards, various factors are applied to all stress calculations .......... Me? I design by the book and apply the book factors, and then multiply by 1.0 to 2.5 AFTER applying all the book factors. We DO hit things and we DO drop things ......... and I REALLY dislike component failure."
Of course, when you're building structures for a race car, it's often difficult to simulate the exact stresses the structure will receive in actual use (and here I'm talking about "simulate" as in create a real physical test for a real part, not computer modeling). This is because the stresses are so many and come from multiple sources. F1 teams like Mclaren and Ferrari have multi-post test rigs that cost millions of dollars, just for this purpose. This is because the more accurately you test, the smaller "fudge factor" is required, and thus the lighter the structure.
But we're talking about a bicycle crankset, and that should be easy to test. If you're planning to build a crankset for weight weenies, then you should by definition have an accurate testing rig which effectively simulates the toughest real-world conditions you can imagine for you parts. A robotic Tom Boonen should put 50k closely monitored miles on the cranks before they ever go to market. Conversely, if you can't afford to thoroughly test the crankset, you should leave a larger safety factor in the structural design.
Now, all these generalized statements I have made could be applied to any product made from metal - say, a Dura-Ace crankset. Parts made from carbon-fiber, and specifically moulded carbon-fiber, require some further consideration. Because the strength of carbon fiber depends on the fibers themselves, the orientation of these fibers is particularly important. In a compression-moulded area (like the spider tabs), generally speaking, a majority of the fibers should be randomly arrayed so the strands propagate in almost every direction, leaving no weak link. A slightly higher percentage of fibers may be oriented in such a way as to reinforce the direction of primary stress, however.
Making certain of this alignment requires a carefully developed manufacturing technique with excellent quality control. In this case, the hypothetical robo-Boonen should have put 50k miles EACH on numerous test cranksets before they ever hit market, and thereafter should continue to test random pieces from each batch. Each piece would be microscopically examined post-test for signs of stress cracks or other indications of potential point failure sources.
Secondly, and this one should be obvious to anyone who understands the basic strengths and weakness of carbon fiber as a structural material - you SHOULD NOT drill holes in your carbon fiber parts. Why? When you drill a hole, you just CUT all the strands from which that carbon structure derived it's strength. Instead, holes need to be MOLDED in - but this is more difficult and thus more expensive.
I could continue to go on at much greater length, but I have a feeling few of you have even read this far, and I'm sure what I've already written will be torn apart by those more knowledgeable than I - indeed, this is welcomed, as it enhances all our knowledge.
As a final statement, I have to say that I feel, morally and ethically, that if a company cannot meet all the above requirements when designing a lightweight carbon part (lightweight implying that they didn't take their computer-modeled design and multiply by 2.5), they should not be producing that component at all. In the US at least, I suspect that any corporate liability lawyer would tell you the same thing. I guess things aren't quite as cut-and-dry in Europe. I never thought I would praise the litigious nature of American society, but at least it makes people think twice (and then think again) about marketing a product that has the potential to fail, and, in failure, cause injury.
And, for those who found this post interesting, I suggest a copy of Carrol Smith's superb Engineer to Win as a starting point, as it has great, simple explanations of structural materials, their uses, and how/why they fail.