It is certainly true that increasing bias/approximation error can lead to decreased variance/estimation error, or even to an efficiently solvable problem rather than an inefficiently solvable problem. But somehow I don't think that's what's usually meant.
I've been reading the recently released book on this
(www.amazon.com/Marshmallow-Test-Mastering-Self-Control/dp/0316230871/)
and it has been interesting & good. I recommend it to anyone
interested in discipline/self-control.