sfo:

isn't this the ackerman effect or some such?

I think this is the result of not enough Ackerman correction at full lock. 

The question is, why does it do it when in reverse, and not going forward (in my experience).