I took a similar approach. When I found a guess that returned a result that didn't make sense (due to the jitter/business of the server) I kept track of it. At the end, I looped through each of those weird results and hammered them until I got something that seemed stable.
I actually had no problems with chunks 1 and 2. Half way through chunk three the server started acting up so I aborted my script and added the modifications that I mentioned above. I tweaked the server and client so that they had a hard coded value for chunks 1 and 2.
It's a bit messy, as my original solution didn't work on the live server and I had to rethink it. My second plan was to reduce it to a handful of potential chunks I could brute force. Once I got it running, though, I realized it was eliminating about half of the chunks on each pass, which left me with the correct one in 7-13 iterations.