In the absence of anyone providing an independent answer, I have researched this topic in much more detail. I believe I now have a good understanding of the issue, and the answer is sufficiently interesting that it seems worth sharing.
(The best reference I found was the information in the documentation for the excellent Bernese GNSS processing software, Vol 1 and Vol 2, especially chapter 10.)
Causes of motion for "fixed" points, when measured in absolute ECEF coordinates
I had assumed that tectonic drift (in the UK at the scope of 25mm/year) was the sole significant cause of change of reference station location when measured in ECEF coordinates (i.e. relative to the Earth's center of mass).
This is not true.
There are multiple causes of motion of "fixed" points, many of which are several orders of magnitude larger than the resolution of modern multi-frequency RTK GNSS positioning.
The major factors are:
- Solid Earth Tides, Solid and Ocean Pole Tide, and Permanent Tides (direct Sun and Moon influence on the Earth's crust, net magnitude ~1m, on timescales from hours to years. Wikipedia entry has some great images of the different elements.)
- Ocean Tidal Loading (impact of mass water movements due to tidal forces)
- Atmospheric Tidal Loading (impact of tidal movement of atmosphere, magnitude a few mm)
- Non-Tidal loadings (including atmospheric pressure loading, range a few cm)
- Other Site Displacements (including post-glacial rebound and post-earthquake relaxation)
The net of this is that a "fixed" point will often move by over 1m during the course of a day. This is why when you post-process raw carrier data to give a PPP solution, tidal corrections have to be applied.
Relative Motion of base station and rover
Having said all of the above, for two points which are within a few Km of each other, and are on the same tectonic plate, the above motions will have near identical extent. This means there will be almost zero change in the relative location of the two points, even though their absolute positions are undergoing significant change.
RTK GNSS calculates the offset of the rover relative to the base station, so this is not impacted by the bulk movement relative to the Earth's center of mass.
This is why I am able to attain repeatable results to better than 10mm.
Impact of conversion into local Geodetic Coordinate Reference System
Whilst use of a specific Geodetic CRS with local datum, leads to a similar zero relative motion of static points in scope, there is complexity when converting from ECEF to local CRS in different epochs.
In most CRS, the local datum is fixed relative to the local tectonic plate.
In my case the UK OSGB36 CRS is fixed relative to ETRS89, which is in turn tied to the main northern European plate. The datum locations for ETRS89, are defined in ECEF coordinates measured in January 89.
Since then, there has been ~85cm of datum drift relative to the original ECEF coordinates. To accurately convert from current epoch, absolute ECEF coordinates to OSGB36, requires adjustment for this drift.
OSGB offer online tooling for this conversion.
If I was using this tooling for my own surveying, then I would need to periodically update the ECEF location of my RTK base station, if I wanted repeatable OSGB36 coordinates for fixed points.
If I didn't, then the OSGB coordinates of current points would show increasing errors relative to past points. This is because their ECEF coordinates would show zero drift (as the delta to the base station doesn't change, and the base station ECEF is fixed) but the OSGB conversion would be compensating for tectonic drift (so you would effectively see the points moving in the opposite direction to the drift).
However, the current mapping software I am using, has a fixed conversion from ECEF to OSCGB36, that doesn't take into account tectonic motion. In this case I should not update my base station ECEF location. If I did, the rover points would see the (correct) change in their average absolute ECEF location, but the fixed conversion in my mapping software to OSGB36 would show this up as motion in the OSGB36 CRS.
Summary answer to original question
Assuming you want "fixed" points in you local CRS to have constant coordinates in that CRS, then:
If your "ECEF to local CRS" conversion compensates for continental drift in the current epoch (relative to the CRS datum epoch), then you should update your advertised RTK base station ECEF location.
If your "ECEF to local CRS" conversion is fixed at a past epoch (so doesn't compensate for continental drift), then you should not update your advertised RTK base station location.