Is this an artifact to do with J2000 and ICRF?
It's an artifact of choosing to use right ascension and declination to describe the Earth's orientation. Right ascension is undefined at a declination of 90°, and is poorly defined at declinations near 90°. The heart of the problem is that the Earth's declination is defined to be 90° at 12 noon TT on 1 Jan 2000. That's exactly where right ascension is undefined. Near that point in time, there is barely any difference in orientation for even a 180° change in Earth's right ascension.
Satellite tracking antennae experience a similar problem when the satellite they are tracking is very close to directly overhead. Most tracking software algorithms avoid abruptly changing azimuth when the tracked satellite is directly overhead because if blindly applied, the tracking algorithm would have the antenna spin wildly. The antenna instead has to spin enough and smoothly enough so as to ensure the tracked satellite remains in view throughout the singularity and will remain in view after the singularity passes. A 180° rotation exactly at zenith is not needed. That 180° azimuth rotation is needed because that is the way tracking antennae work, but it can be done slowly. The spacecraft? It simply passed overhead. Azimuth and elevation work nicely away from the vertical, not so nicely near the vertical.
Another way to look at this is that using right ascension and declination to describe the orientation of the Earth's rotation axis is not the best choice. Yet another way to look at it: This is an incredibly bad choice because that choice passes through a singularity at noon TT on 1 Jan 2000. Nothing physical happened to the Earth at noon TT on 1 Jan 2000.