This is a long read, so my apologies for that. But this thread is so full of speculation and errors that I thought I'd just contribute what little I can do that's impartial and factual.
First, I have no dog in this fight. I'm not affiliated with Shimano, a bike shop, or anybody else in the industry. So I have no reason to cover up or to foment a storm.
Second, as far as credentials go, my career (and my doctorate) have been focused on high volume manufacture of precision composite devices for medical and military purposes (both areas with high regulatory oversight). I've also worked in the past with a couple ultra high end automobile manufacturers and a couple carbon fiber bike component manufacturers to do manufacturing design and failure prediction.
Third, any time a bike equipment manufacturer encounters a failure (typically a report from a bike shop and usually a matter of replacing an item), the manufacturer has to report it to CPSC. If any number of problems arise (the number determined statistically based on manufacturing numbers reported by the manufacturers to CPSC), they escalate the problem and begin to investigate it. They are watching for long-run trends, for bad manufacturing runs (cluster failures), and for consistent patterns of failures. You can FOI most of this information but it's usually processed for release and won't tell you much -- only attorneys find it useful.
Now on to the failures documented here. 4914's failures were documented in two threads on this forum (and are the only postings he's made on WW). As his forum name says, he sees himself as a former track sprinter. Now 4914 shows three photos of failures. The first one involves failure of the outside cog of the three (which has its own set of rivets); the other two superficially involve failure of the other two cogs (which are joined by their own quasi-independent set of rivets) (but let's get back to this in a moment). Let's look first at how this group of cogs is engineered. The outer cog is supported independently by its own rivets, and the back of the rivets additional bear on the shoulder of a fairly substantial pillar. This is the only actual rivet failure I've seen. In the other two cases, there's no clear way to know which cog of the inner two was in use when the spider actually fractured, but since they are joined and the rivets still appear to be intact, they had to break off as an integrated unit. If you look at one of these spiders, you'll see that the single outside cog is mounted towards the rear (counterclockwise) side of the spider, so it actually has most of the spider to break through. The two in the rear actually bear against the spider plus the rivet for the single outside cog. That's good engineering to make this unit extremely strong. (This is also why the rearmost two cogs aren't mounted on a carbon fiber spider. Plus, if on a low flange hub like most road hubs today, if the chain derails into the spokes it will drop down against the spider and a carbon spider can't handle that as well as aluminum can.)
Now, the question was raised of whether carbon fiber was strong enough to begin with. If carbon was used in a unit like this on its own, then the answer would probably be no. But the parts of this cassette unit act in an integrated way. The spider gains immense strength from being pinned to two different sets of cogs -- you would typically have to snap a cog to weaken the spider (or have a rivet failure). Now go back for a second to the observation above about how one rivet bears against the other in the spider. If you look at #2, you can see that the front cog has actually failed and that allowed the rear cog to collapse shortly after. In #3, the rivets of the outside cog are still at least partially enclosed within the material of the spider, so they never dislocated -- this was a failure of the inside ones. And the inside ones actually ripped away the carbon surrounding the outside-cog rivets as well. This is a very improbable failure.
There's a thread cited on an Australian forum that refers to someone who broke two 9000 cassettes back to back. No photos, so no way to assess what happened. There's also a poor Twitter photo of Craig Bennett's cassette having a failure, and in that case it isn't clear what happened -- whether the chain simple derailed behind the cassette and collapsed the cluster, or whether it suffered a failure of the inside or the middle spider is unclear. However, the tooth configurations and the continuous drilling around the smallest cog are inconsistent with a DA9000 cassette anyway. It might be Ultegra or possibly something nonstandard, but it isn't a 25 or a 28 Dura Ace 9000 inner cog. So we can't deal with those instances in any analytic way. So, back to 4914's three failures.
There's no evident damage on the back side of the inside cog, so it doesn't appear that something was happening on the inside that snapped the cogs outwards. However, look at the splines that mate with the freehub. All three cassettes, for being brand new, show significant fraying of the splines. Now since the center of the spider is intact in all three cases, there is no way that the spline could lift up and pop over the freehub splines, at least so it could fray a few of the splines and not others. That damage could only come if the spider were flexed hard inwards or outwards. It would be helpful to see both sides of all three spiders, but assuming there's no telltale scoring there, I'm starting to think that there's either something wrong with the chain (wrong chain sizing or defective links?) or something wrong with the setup of the bike so that the chain can lever against the cogs in the middle cluster. All three failures could be explained by the chain (or possibly the rear shifter acting on the chain) prying the middle cog of the spider until the spider gives way. The spider isn't engineered to be strong in that direction and shouldn't need to be, but if a chain of the wrong width was used or if the rear derailleur was adjusted (or chain length was wrong) so that it jammed the chain and snapped it sideways, it could explain not only one failure but also why the bike kept destroying cassettes.
Any time you have one cassette failure, it's a very rare event. When one bike causes multiple successive failures and does so almost immediately, you have to look at the bike and how it's set up. Looking in detail at the damage to the cassette, everything is consistent with a prying action from side to side, not with a rotating load (for which it's engineered). Once the damage begins, when the chain tries to run over the cog in question or an adjacent cog, it jams and tears everything apart. The torque vector applied by the speed of the bike is far greater than what any rider can accomplish, and greater than what the cassette is engineered for, so it can tear the cassette apart once the cassette has been damaged. And the source of that original damage? My guess -- and this is only a reasoned guess at this point -- is that something about the bike caused a sideways deflection to damage the spider or cogs on it. That would explain the recurring failures and why the cassette is otherwise basically reliable for almost everyone else on this forum and elsewhere.
The creaking is somewhat of a bogus issue. We all know that carbon against metal likes to creak. That it happens here is probably irrelevant. If it's relevant, it's only because creaking might be accentuated if damage had already begun. But these cassettes do creak and do need retightening and lubrication. I wouldn't make creaking a direct correlation with failure.
A while back we collected almost four hundred hubs that had undergone flange failure due to radial spoking, so we could understand whether radial spoking was really an issue or not. We did similar studies on tubulars that had rolled off rims -- almost two hundred tubulars. That kind of failure analysis is what really tells you what happens on the bike. Send these cassettes to a common repository and let's start looking at them. I think we'll find a common thread that isn't a pure design failure. Until then, I'd keep riding my 9000 cassettes and not worry about them. I'd change chain, chain length, and rear derailleur setup (especially b-screw) and see if we could get rid of the problem. It's a new bike, remember. So I'd start there.
Just my $0.02. But I'm trying to stick to the facts and to looking carefully at the photos that were provided. Probably just time to move on.