The byte1 /byte2 line is in effect "A0", selecting which of the input latches is written to when /WR1 is activated.
As those have to be loaded during separate instruction cycles, if they were directly linked to the DAC, it would have part of the new value and part of the old value between those two cycles.
The second latch can load all 12 bits at the same time, when the /XFER and /WR2 signals are activated.
The full address decode to set the physical addresses the DAC setup operates at is not shown.
Note that this setup could be simplified, with just one latch and the other data lines straight to the DAC register; load the latch then write the DAC register, which would take the latched data for one half and the new data for the other.
That's how 16 bit peripheral transfers are done in some MCUs; eg. write a holding register for the high half of a word, then write the low half to another register, which also transfer the stored high byte.