The main issue arises when a split or join of a leaf drives a structural change that propagates up the tree. If we ignore this, then we can split out a workload based on the keys in the root node of the index and apply the updates independently. Therefore one possibility would be to allow the nodes to overflow or underflow the target branching temporarily, finish the updates distributed across those key-ranges of the index, and the "fix up" the index. Special care would be required to ensure that the range counts are also valid after such a fix up. There will be interactions with the write retention queue eviction strategy when considering any such change in the basic mutation strategy of the indices.