Because while your OS might be split-copying your database blocks during the bacup operation of your datafile.

Suppose your database block size is 8K. And let's suppose your OS does read 4K OS blocks at a time while copying a file from one location to another. Let's say the OS reads first half of your database block Bx at time T0 and starts copying it to the backup location. At time T1, the database writer changes some data located in the second half of your database block Bx. At time T2 the OS finishes copying the first half of your database block and starts copying the second half of yor database block Bx. So in the backup file, the first and the second half of your database block are from diffrent times - in other words, your database block Bx is inconsistent. Yet it represent the foundation for your media recovery. So if Oracle is to deal with this block inconsistency, it has no other choice as to record *the intire database block* in the archive log file *whe the first change occurs in this block* if the file is in backup mode.

Note however that the entire block must be archived only when *the first* change to the block is made after the file has been put in a backup mode. Any further changes to that particular block will cause only vector changes to be recorded in archlogs, just as during the normal operation of the database.