Thanks, more data is always nice to have ;o).
(In particular, it confirms that the forced time adjustment on resume appears to be discarding nanosecond precision (and/or it's just a timing artifact, but the bottom line is essentially similar: subsecond drift, either backwards or forwards) on those older kernels, which effectively translates into time going ever so slightly backwards, so I'm glad I actually kept the discontinuity check (I was considering relaxing it severely, because the whole thing forces us to run a couple syscalls just to correct that small subsecond drift after every page turn), because it's technically more "correct" ^^).
Timekeeping is fun. It'll be even more fun in 2038 when all of this will self-destruct

.