There were some general guidelines I recall posting a year ago that a guide scope FL should be 1/3rd the imaging scope. However this did not consider many factors such as differences between image scales of the imaging chip and guider etc. I also believe advances is guiding algorithms has lead to greater accuracy. Most algorithms calculate on the guide star centroid with sub arc/sec accuracy. I'm aware of a 3300mm (14.5" RC) being successfully guided by a Tak Sky90 operating at ~400mm, so don't believe the tolerances are so high as once believed. I've tested the image scale theory myself with my own basic rig - imaged at 3.5 arcsec/pixel and guided at 5.7 arcsec/pixel with no problems at all. Not hugely different image scales, but different. The common issue here is flexure. Regardless of the guide scope, if flexure exists, its going to be difficult to guide accurately. On axis guiding has its benefits, but also short falls - finding guide stars in a small FOV is one of them.
This link maybe of use in your quest -
http://www.wilmslowastro.com/tips/autoguiding.htm
I'd recommend speaking with Fred (Bassnut) as he's done some work in this area before with his own long focal length rig. He'll probably chime in. A few others are pushing the 2000mm FL barrier too.