When a layered cursor is cached, it simply closes its constituent cursors. While these "sub-cursors" are cached and will be "uncached" when the layered cursor is reopened, the reopen call has some overhead (looking up the URI, parsing any config string). This can be sped up by using internal cache and reopen calls on a cursor. That is, cache the cursor and crucially retain the pointer to the cursor. Then when the layered cursor is reopened, we call the internal reopen function using each subcursor.