> is it worthwhile to add such optimizations for without-rowid tables too? Such an "optimization" actually slows things down in the common case. If we had a way to determine when the optimization would benefit, then it would be worth adding. But I don't yet have an algorithm to do that.