A query with multiple range predicates that scans a compound index can end up doing many small seeks--in the worst case it can do one seek per index entry, which is slower than scanning those same entries would be.

Let's try calling SortedDataInterface::next(), optimistically. If we are doing few big seeks, then this shouldn't add much overhead, and if we are doing many tiny seeks it could avoid the overhead of SortedDataInterface::seek().