This has been discussed here before: standing people pack tighter than walking people, therefore all standing ends up with more throughput, at the expense of lower individual latency (for the would-be walkers).

But you only need that optimization when escalators are at full utilization. In many cities, escalator discipline is low enough that the walk/stand lane pattern deteriorates when the escalator approaches maximum capacity. The minority who don't get it and rudely block the walkers accidentally apply the throughput optimization. London seems to lack a reliable distribution of rude blockers who would accidentally optimize for throughput when necessary, so they need to search for an explicit solution.