Reading a node state in a past revision can become expensive when the change history in the previous documents have overlapping changes. In this case, the changes in the previous documents must be merge sorted to find the correct value for the properties on the node. The more overlapping ranges there are, the more sorting is needed.
There is a prominent node in the repository that seems to create quite many of those previous documents. The /:async nodes gets frequent updates and is therefore split on a regular basis. Because the properties on this node are not all updated at the same time, it is quite likely that previous document ranges overlap.