Currently, get operations just fetch from disk, whereas it is probably faster to optimistically traverse through the in-memory tree (falling back to disk as soon as we hit a pruned link).
Just tried this out in f1548f5. It has no effect on pruned trees (as expected), but for a cached tree on my i9900K machine the reads go up to a cool 2.37M / s.
Currently,
get
operations just fetch from disk, whereas it is probably faster to optimistically traverse through the in-memory tree (falling back to disk as soon as we hit a pruned link).