Filtering the documents being specified in a feed based on whatever criterion is currently not part of the RSS connector. The only means of filtering is by including or excluding the feed itself.
So it sounds like what you need for this case is NOT to understand a proxy.pac file, but rather to permit discovered URLs to be filtered in some way. Will being able to filter based on regular expressions run against a document URL be sufficient? The web connector uses this strategy, but it seems to me like it would be problematic in an RSS situation. Presumably the mix of links will be changing all the time, as the feeds are regenerated; you might possibly be able to decide via a regexp whether a link was internal or not, but it will be cumbersome to manage this I think.
The alternative is to generate the feeds without the documents that you don't want.
Please let me know how you want to proceed.