I have recently developed a staking plan based on dutching. I've back tested it against all events in 2015 and 2016 (~20k) with an 80% hit rate - which I think I can improve on.
The problem I have is I work full time so deploying it is impossible right now.
Testing is key here and I can't stress that enough, I had situations where the first 100 races was smashing it but then it would go right down the tubes so you need to test it against as much data as possible, I tested mine against 20k historical races over 2 years. Build, test, tune - rinse and repeat.