They haven't gotten to the good part yet. Linux uses all available free memory for disk caching. Pages which are read cache can be freed when memory is needed. The problem is that there are cases when something needs memory, and the cache space can't be freed due to some transient locking situation. This results in a kernel panic.

This has never really been fixed, just mitigated by trying to keep enough memory free that the kernel doesn't get itself into that corner case.

I've not heard of this and searching doesn't seem to throw up anything relevant. Do you have more details? Maybe a link to the LKML discussion?