By "some time now" he must mean 3 months, since it was trivially bypassed by using mmap instead of read/write for the 6 months prior to that (since 2.6.26 when it was CONFIG_NONPROMISC_DEVMEM). See: https://bugzilla.redhat.com/show_bug.cgi?id=460857
As you can see, that vulnerability was only found and fixed because it was being actively exploited in the wild by a rootkit (so much for Linus' "if we don't tell the bad guys about the bugs, they'll never find them").
Compare that to the 7 years grsecurity has protected against this. Not only that, but as the PaX team pointed out in the other thread, the current code is still wrong.
It seems to be a recurring problem: security concepts that get adopted from PaX/grsecurity are never implemented correctly on the first or second try. Remember the NULL pointer dereference "protection" which was bypassable for 6 months via expand_stack and other privately disclosed methods? Or the weak ASLR added since 2.6.12 (that's 4 years ago) where only last month did it become public that the "randomized" bases were trivially obtainable locally via /proc (and there are still methods that aren't fixed). grsecurity had that closed for 7 years as well.