As for IBM my argument is that the Xbox One follows in the footsteps of the 360 design (eDRAM) and then shrink (first APU) and that IBM was heavily involved in that design.
I also have heard that a number of key IBM guys moved to MS for the Xbox One SOC effort.
Also, the common platform alliance involves IBM.
The common platform alliance probably does what Intel does: Runs massive SRAM chips on a new process. (One of the first circuits they run, pretty early in the process development.) It is a key way they characterize yield and performance.
But keep in mind that if it is 6T or 8T then it would seem to be rather similar to the L1 and L2 which I am pretty sure is not having issues. I would be really shocked if the APU portion had issues with the L1 and L2. All of the recent AMD large CPUs have a good chunk of cache on them and I have heard zero about issues there.
Your counter arguments are certainly true as those things did occur. So yes, engineering mistakes happen all the time. But this rumor does not smell right to me at all.
Take a look at this Intel tech PDF:
http://download.intel.com/pressroom/kits/events/idffall_2009/pdfs/IDF_MBohr_Briefing.pdf
What I am saying is strongly rooted in engineering understanding.
That PDF should help with the SNR on this issue. They know LOTS about the SRAM capability long before any CPU manufacturing starts in a process. And those are LARGE SRAMs.