We have had problems like this and we couldn't locate the problem neither. After searching for a long time we found out our sqlserver was overloaded (because we logged to many info's to the server). Now we log to a txt file and everything works fine.
Our flex application is a test tool which 350 candidates (at average) take at the same time. The load of the webserver and sqlserver is pretty high because every 30 seconds the state of a flex application is saved to the server (state = current exercise the candidate is working on).
We haven't had any problems, so I'm guessing your problem isn't the flex application. Maybe it's the network?
We do work with WebORB and amf remote calls. We also have a retrieveBlob.aspx handler used to fetch media. On average, when 350 candidates start there session, some 3500 media files are fetched at the same time (with caching on the server). In this case we haven't had any problems so far.
Another thing we implemented is a fallback system where if some remote call fails, flex calls it again with the same parameters. If it fails again, then an error message is shown to the candidate saying it needs to restart the application (and a resume is done).
Flex automatically retrying the post or get is something I haven't seen yet in our application.
Which server are you using? (we use Flex-Weborb-asp.net-sqlserver).