The diff is averaged over 2016 blocks - so a drop right before wouldn't really impact that much.
Well to do the math on that

2016 blocks - 1% is 20.16 blocks - 0.1% is 2.016 blocks
- so with an extra 20 minutes in a single block (or 2 missing average length blocks) you can reduce the diff change by almost 0.1%
Edit: and the diff change was:
[2018-02-07 10:04:08.448+11] workinfo_add(): DIFF CHANGE: hi=508032 delta=10.43% new=2874674234415.9 prev=2603077300218.6
Edit2: and totally randomly, but slightly related

the last 4 blocks since the diff change have averaged 20 minutes each
... and each has been pretty close to 20 minutes - 19min, 17min, 19min, 25min