Some nice info described in a LinkedIn HFT group post. Scott describes how hit bids/offers are provided with an ACK packet before the prints or order book is updated. This means if you can get a sacrificial lot early in the queue and the rest of your order slightly behind that lot, then you can decide to leave or pull the rest of your order before other traders have the same information as you. Further posts in the thread say the CME is actively preventing this, but there are work arounds.
If there is 250 (size) on the bid (ES for example); with an institutional data feed you can see how many individual orders comprise those 250. You will normally see an average size per order of between 5 and 10; that means there are some larger orders mixed in with many 1 lots.
When an order fills, the owner of the order gets an ack from the exchange before the trade prints and well before (in HFT terms) the book size is updated. The owner of the order knows the price traded before everyone else.
So as you correctly state, the owner of the order pulls the rest of their (shredded orders). They can do this because they knew how many orders traded in front of that fill and if that is sufficient to give a statistical probability the level is going to trade through, they pull the orders.
This process can be repeated many times on the same price level if when placing the orders, a time delay is introduced between each placement (or possibly between a small group of orders). You may be 230 out of 250, 215 out of 250, 190 out of 250, etc. If you get to an order in your order stack that means too many orders have traded at the level for the risk profile the algo is using, the algo cancels the remaining orders.