Aren't splay trees terrible because they turn all reads into writes? (Which means the cache lines bounce between readers, instead of being shared as they would be with pure reads.)
Yeah, having reads rebalancing the tree in a multithreaded subsystem is probably not optimal. Might be that the Splay-trees have outstayed their welcome :)
FreeBSD’s tree.h used to have, iirc, both RB- and Splay-trees.
