I think you're giving Cat3 riders somewhat too much credit. I think a 17ish 5k is more competitive in the running world by a fair amount than the Cat3 cycling denomination, but that's just a guess based on my ability to hang with a good number of Cat3s and my completely inability to run at all at any distance with 17:xx 5k runners.

I also don't think you're correct about marathoners having a tough time winning Cat3 races. I think as a whole, you'd find more potential in that marathon trained group due to their experience with endurance sports and built up endurance compared to their nonmarathon peers. This whole 'slow twitch' thing limiting performance in cycling is way overhyped, as even in crits where you're doing repeated anaerobic sprints, your endurance is still by far the dominant factor. It's also definitely not true that marathon selects only people with slow-twitch fibers; for the amateurs, it's pretty much a mix of all of them, and test of current elite marathon runners show a real trend toward fast-twitch ratios, meaning they're speedy guys who can run long.

If you look at people's running race results from 5k - marathon, as long as they're trained for the marathon, their performance across race distances are remarkably consistent - it forms the basis for Daniels' VDOT tables and many other similar tables which have very good pedigrees. It's not like guys with fast twitch fibers don't have these Vdot tables apply to them.

Here's Daniels' VO2 (aka Vdot) tables:
http://www.moorestownxc.org/Training/JackDaniels.html

At a 17:03 5k, you're running a 2:43 marathon. That's faster than the top 0.1% of marathoners in the country, and still in the very top few percent at the qualifier Boston Marathon. I'm pretty sure that's much more competitive than a typical Cat3 performance.