No, it is not moot for that reason. In order to use pointers for in-memory table row references and rowids with b-tree traversal for on-disk table row references, there would have to be significant duplication of functionality to provide row reference creation and use, in portions of code which, today, have no dependence upon the physical form of storage. The in-memory and on-disk distinction is only made at the VFS level; the code which produces the file-format (even in memory) is insulated from how the block I/O is implemented. That would cease to be true if this putative optimization were to be effected in the non-VFS portions of the code.