Probably because the guy who wrote the function tried a few values and ended up with
1275 as the closest number of loops to make 1 millisecond.
It sounds like an odd value, anyway. 11.0592 MHZ divided by 1275 makes 8674.
With a value very close to 9000 for instance, I would guess that 1 loop takes 9 clocks.
By the way, I suppose you took the code as is from another implementation that was
working with another crystal, right?
Writing a raw value (like this 1275) is just a sign of a poorly written code.
What you should do is:
#define LOOPS_PER_MS 1275 // Number of loops per millisecond
and later for(j = 0 ; j < LOOPS_PER_MS ; ++j)
Doing this, you will have no trouble to understand what you wrote even
a few months later, and on top of that other people will also understand
what you wrote.