In the presentation posted here there was a paper citation that compared in terms of cache the model used (parallel depth first scheduling) over work stealing, but I didn't have time to read it yet.

https://www.cs.cmu.edu/%7Echensm/papers/ConstructiveCacheSha...