1) I think it's 350mcs across the board except in the case of firms which conduct suspect activity (canceling orders frequently for example). Adding the delay institutes a level playing field across anyone trading on the exchange. The basis of HFT is leveraging faster connection speeds to gain insight into other's trading strategies and exploiting those strategies, all before the other firm's trade reach the exchange. This is a HIGHLY simplistic explanation dealing with one form of arbitrage and given for brevity on the subject.
2) EDIT: the other guy explained this better!
To your point about 350mcs being absurd to argue over, HFT firms manage regular trading speeds in NANOSECONDS. Look at 350mcs in those terms (350,000 nanoseconds) and it's not such a small number anymore.