> So, this issue in some way related to PaX changes between 2.6.27-r8 and > 2.6.28-r7.
Maybe; there are also a lot of core kernel changes between those versions, any of which could interact with the hardened patchset in a negative manner. Given that identical systems with identical configurations work, I find the specific, physical machine more suspect than something with PaX. The memory management changes it provides in combination with the memory management changes introduced in 2.6.28 could well exacerbate existing issues in a RAM module that weren't being triggered previously.
