The move instruction does not use hub memory. It's strictly a cog memory instruction and takes 4 clock cycles like most of them. RDxxxx and WRxxxx are the only instructions that move data between cog memory and hub memory (other than COGNEW/COGINIT).
You'd have to look at the Spin interpreter source to verify, but I believe BYTEMOVE uses RD/WRBYTE, WORDMOVE uses RD/WRWORD, and LONGMOVE uses RD/WRLONG. These are not indivisible operations though since Spin uses no locks internally and the Spin interpreter has to do a RDLONG, then an ADD, then a WRLONG, then an ADD, then probably a DJNZ for the count. This would take at least 2 hub cycles per transfer.
Comments
It has to do with saving memory to a fram without corrupting high and low order bytes in a word or long.