The most visible difference between the two is that the
assembly version uses a function which is (presumably)
tuned to the system speed to give an accurate delay,
and the C version issues a fixed number of NOPs
independent of the CPU speed. My guess is that the
LongDelay isn't "Long" enough to be detected by
whatever's connected to P0/1. I.e. either
1) The compiler is optimizing out the NOPs (unlikely
but easy to check) or
2) Your instruction clock is so fast that the NOPs
plus the loop arithmetic doesn't add up to a
detectable delay. (E.g. with a 0.5us instruction
cycle this is maybe one-eighth of a second,
which is probably not detectable e.g. by eye on an
LED.)

You might want to peek at the assembly code that the
compiler is generating.

Does your C library have a (tuned) delay function?