I think the standard method of dealing with this is to have your users upload the items to the local web server and then have some sort of async task or cron job responsible for moving the files to S3 later on.
Of course this means you'll have to keep track of where the uploaded file is currently located, but if you ever want to scale to multiple web servers you'll need to do that anyway.
Yea - this is how I've been doing it. The server that gets the upload does some processing and then moves the content over to S3 and adds an entry to the database that references it. It also forces you to write stateless code which is nice bonus.
Of course this means you'll have to keep track of where the uploaded file is currently located, but if you ever want to scale to multiple web servers you'll need to do that anyway.