The only real way to test a HFT system is to use real orders on low volume then progressively add volume to determine at what point size becomes a problem. This is by far the simplest, fastest and best method. Anything else is just guessing. If you can afford real money tests, you can't afford to play in the HFT space. It's a big boys game.
Most crucial issue is to have the right data. If you're reasonably good at coding it's not that hard to build a properly functioning orderbook simulator to test your strategies on. Yes - modeling queue position is crucial but I found you can get very far and accurate with very heuristic level queue models. Obviously your simulator will have to be able to deal with and model the latency you're experiencing in live trading as it'll affect your queue position, will cause limit orders to cross unintentionally, lead to rejected cancellations and the like. I have experience with futures only so don't know if equities are harder or trickier to get right. Provided you can code, I'd never go with a third party product, as I dont know what their assumptions are and how they handle certain details that might make a big difference in terms of tracking error.
+1 Just turn it on and let it run. You need to budget a few thousand to draw down initially while you test/tweak but that's the cost of admission in the real HFT game. If you are trying to backtest fills in a low-latency space you aren't trading HFT currently. This is one of those "if you have to ask you aren't there yet" type questions.