Out of curiosity, were you working with Yara 2.0 when you were developing? It is several orders of magnitude faster than previous versions.
To your question, I would be interested in this effort but before diving in would like some time to familiarize myself more with Bro development. I will be at this years BroCon in pursuit of that goal and would welcome further collaboration toward this end 
Ideally, what I would love to see is a way to take actions on alerts generated by some kind of ‘Files::ANALYZER_YARA’. So say if I have a ZIP file for example and a Yara rule to detect a ZIP. I think it would be very valuable for someone to not only just trigger on that, but then invoke an event that decompresses the ZIP and feeds the contents through the same scanning engine. Now replace ZIP files with a known crypter/obfuscation or something else and you can perhaps start to see the power and possibilities that begin to unfold 
Full disclosure time…:
I am a malware reverse engineer by trade. When I RE a binary I can tell my customers (the analysts) a lot about it, however, their ability to take action on the intelligence I give them is oftentimes limited by their capabilities / security posture as an organization.
Enter Bro, with a modular framework, I look to this as a means to make the observables I gain from my RE efforts as more valuable actionable intelligence for my team. By implementing this modular ‘take action on X’ mentality with respect to Bro and Yara, my signatures get more milage, as well as my observables on how certain crypters/encodings can be defeated.
Imagine this, I have a signature for shellcode that decrypts a PE in a certain way always at a certain offset. My Yara rule hits on this signature and triggers an event that unmaskes the binary as well, out pops the dropper, that is scanned again, and hits on the signature I created for the dropper, etc, etc…
So I’ve automated analysis that is usually done by someone more experienced on the command line. Not only that, but now the analyst knows more about what they are dealing with which directly informs IR/Intel efforts.
Hope that helps paint the picture a little more 