I was hoping this would be about parallel pointer chasing.

When you don’t a priori have ‘processing power at each node’, you can still parallelise by tracing speculatively from every location in memory at once, and then throwing out all the results you don’t end up reaching from the root.