The loop at Welshpool is not a "special case", it is just that, a long loop.
Sorry, but not true. It has also been split into several "block sections" (or whatever they are called in ERTMS terms), so that theoretically you can have two or more trains following one another within each loop line. Unlike the standard loops, which only have one block section within each loop. Having said that, I had forgotten that Newtown was also made into a dynamic loop, so Welshpool is not quite so "special case".
I can only assume that the reasoning for the restriction on allowing two trains to enter a passing loop simultaneously is one based on safety rather than a quirk in the signalling system? Could it be argued hypothetically that if the path is successfully set for two trains to enter from each end of the loop, they could in theory enter at the same time with confirmation of the set path?
At most of the loops, with ERTMs it is not possible to set for two trains to enter each loop simultaneously. Whereas previously, at the same loops, it was possible with RETB. The reason is that ERTMS requires an overlap beyond the end-of-loop marker/stop board, in case a train over-runs the board, in the same way as overlaps are provided beyond conventional signals. This overlap extends through the loop points beyond the board, so blocking any move being made through those points. So it is not possible to admit say a down train until the up train has been brought to a stand in the loop. Once the train is at a stand, then the overlap through the points ahead is no longer needed.
With the old mechanical signalling, most of the loops had trap points at the end of each loop, to divert an over-running train. Where both loop lines had trap points, it was therefore possible to admit two trains simultaneously. These trap points were removed when the line was resignalled with RETB, as there was no way for the RETB to control them.
With RETB, the loop facing points were worked by hydrostatic self-restoring point units, which imposed quite a severe speed restriction on trains entering each loop. Due to the low speed at which trains entered each loop, only a minimal overlap was provided beyond the end-of-loop stop boards, which did not impinge on the loop points beyond (I have a feeling that, even with the low speed, a derogation against standards was needed). So with RETB, it was still possible for both up and down trains to enter simultaneously.
When the line was resignalled with ERTMS, longer overlaps to current standards were needed at the ends of each loop. This may or may not have been due in part to the loop points being converted to clamp-locks operated by the ERTMS (the hydrostatic units were life-expired), so the speed restrictions imposed by the hydrostatic units were no longer necessary.
Re-instating the trap points, operated by the ERTMS, would be one way of re-instating the simultaneous arrivals. However, this would cost money, the extra equipment would be liable to failures with consequent effects on service reliability, and would involve extra overheads in on-going maintenance. Extending the loops beyond the stop boards to provide adequate overlap distance would be another way (as was done at Welshpool and Newtown when they were extended to make them into dynamic loops). Again, this would be costly.
Although promises were made at the time to solve the issue, I suspect that when the costs involved were compared with the time that would be saved, it was decided it wasn't worth it, and to live with the delays.