From the quick glance over the source code, it seems that this implementation has this assumption too. (I may be wrong, though - it is possible I've missed something.)

On the other hand, Redis has "VM" thing, which allows larger-than-memory databases by swapping less frequently used data to disk.