try this:

db_autopwn -p -r -e

"-r" is telling it to reverse connection back to you.

or better still use"

db_autopwn -p -t -r -e

perhaps its crashing the service by the time it tries the msdns exploit. 

I always considered db_autopwn to be a best effort attack where you are throwing everything you can at a box and hoping one of the things makes it.  If you have enough information to know that an attack should succeed, doing that one directly is preferred.  If you are verifying vulnerabilities and not just trying to get into a box, I'd say if at first you fail, try at least another time or two as sometimes the attacks don't work on the first shot, especially if you've already launched an attack on that service before. Not to say I don't use autopwn on occasion, but if I have a vulnerability scan with CVEs that I can load in there, I'm more likely to use it just to list out what exploits autopwn would use and then use each of them manually on my own so I can control better the options for launch.  Anyway, just my thoughts.

Haha... unless you have a good insurance policy for when you crash something completely and start costing the business money

It didn't sound like he was doing a pen test (gosh I hope not) but instead trying to learn more about Metasploit and autopwn.  Autopwn just isn't extremely flexible, but the thing about it that I think is nice is the fact it will aggregate vulnerability reports from nessus or something else if you can get it into the database and then look up CVE's for you to tell you what exploits match the CVE's in a hurry.  Using autopwn for actual attacking (on test machines you own I'm sure) is going to give you false negatives for certain types of things as the service may not have recovered from the first set of junk you sent it for the attack that should work to actually work.