Timestamp ordering of the initial SH deposit transaction. In case of two peers applying to be an SH in the same nanosecond, lexicographical ordering of a hash of their deposit transactions decides. I believe this could not be an issue because only the same guy could realistically synchronize the timestamps with that precision, so he can only decide the distribution of a continuous sequence of IDs to his own SHs. But if you can exploit that in your attack, you're welcome.

To set the numbers straight, suppose that the honest SH count is a Poisson process such that there is one SH registration expected per hour.