What's the purpose of the CIL nop opcode?

NOPs serve several purposes:

• They allow the debugger to place a breakpoint on a line even if it is combined with others in the generated code.
• It allows the loader to patch a jump with a different-sized target offset.
• It allows a block of code to be aligned at a particular boundary, which can be good for caching.
• It allows for incremental linking to overwrite chunks of code with a call to a new section without having to worry about the overall function changing size.

In one processor I worked for recently (for four years) NOP was used to make sure the previous operation finished before the next operation was started. For instance:

load value to register (takes 8 cycles) nop 8 add 1 to register

This made sure register had the correct value before the add operation.

Another use was to fill in execution units, such as the interrupt vectors which had to be a certain size (32 bytes) because address for vector0 was, say 0, for vector 1 0x20 and so on, so the compiler put NOPs in there if needed.

One classic use for them is so that your debugger can always associate a source-code line with an IL instruction.

I've also seen NOPs in code that modifies itself to obfuscate what it does as a placeholder (veeery old copy protection).

As ddaa said, nops let you account for variance in the stack, so that when you overwrite the return address it jumps to the nop sled (a lot of nops in a row) and then hits the executable code correctly, rather than jumping to some byte in the instruction that isn't the beginning.