Last update on this, at least for today:
It dropped off pretty fast:
A little later (01:00-02:00) it was a little over 220 servers off by a second (plus/minus 50ms).
02-03: 164 servers.
05-06 (now): 66
There are a few dozen more that are off by a significant amount but less than a second. At a glance they look like slowly recovering.
All are excluded from the DNS records, but old clients might be using them of course.
I didn’t look super carefully, but I also saw a few servers that were correct at midnight UTC, but were off by a second for a brief period hours later.
If there’s any conclusion then I think it is that handling leap seconds is still pretty much broken. Obviously not all operators of servers in the NTP Pool take the same care or interest or have the same knowledge of all this, but my expectation is that the overall NTP installations aren’t doing better than this.
It’d be interesting to break this down by server software and versions, but we don’t have that information in the NTP Pool system.