As you suspect, fallback only applies if the primary source is not available at all. If it get part-way through a pull task and then fails the whole pull task is deemed as failed and ePO keeps everything at the previous version to maintain data integrity in the repository.
You would need to look to the server logs for the possible causes of the failure over http.
Thank you for the help.
The logs don't give a good picture of what happened other than couldn't find a particular file. Not sure if it's the Akamai load balancing that McAfee uses where it's hitting a server that hasn't gotten all the updates yet or not.
I'll have to submit an FMR then for the fallback feature as I think that would be better if it worked that way. Or at the very least, some sort of automated retry if it failed or something.
ePO already does automated retries in the event of failure, in your case the error must have been more persistent.
A source server that is in the process of updating should not be available to customers either.
Please submit an FMR about using the fallback in the way described though, it sounds like a good idea.