That's also true. :-) You probably mean the aforementioned `OLSet` / `CompactTable` which do that. That can get a lot of space savings since the sparse hash table part can be an array of much smaller objects - just the index.
Those smaller objects should really become a little larger `(index, truncatedHashCode)`. You only need enough bits of the hash code to be a good comparison prefix to make loading from the `seq[(K,V)]` for a `K` compare only happen when it is very likely necessary - at the end of a "hit" and never for a "miss". Without such a trick, every probe must load from both arrays. Not sure when I'll get to that.
