To avoid unnecessary overhead, MULTIBUF always writes (or reads) a couple of data blocks at once to (from) tape. Making this value large slightly reduces CPU usage, but makes it waste a bit of space at the end of every tape. I suggest using a value that is both significantly lower (when multiplied with the block size) than your tape's capacity as well as much bigger than the relatively small block size. The standard value is 100, giving you (with a block size of 1024) a total buffer size of 100 Kbytes. So you can be sure to waste no more than 100 K per tape and still reduce CPU overhead by two orders of magnitude.