Yes, it's true, my projects are not "burning" of speed and processor business. But, as I most use 9600 Bd, it takes 1ms for 1 byte. If the instructions are 1-3us, let's say that I miss 500 instructions. At 10 times baud-rate - only 50. When you count stack operations to direct to- and from the interrupt routine, it's also a waste of time. You must organize the routine to fetch the next byte to be sent, and so on.

Another possibility: after loading SBUF I call the CRC routine, so that time is not wasted. After return, there is classical - jnb TI,$ to fill up the rest of time (which is not long, CRC calculation takes much)

Regards, Slobodan