Predicting the next session from Dynamic ratio weights

I've not had any success finding information on how the sessions are allocated when using a dynamic ratio method.

If I have a simple network of 2 nodes with different weights (as calculated from a SNMP DCA monitor) is the allocation of the next incoming session predictable with dynamic ratio balancing? (In testing it appears random whether my next session is assigned to the higher or lower weighted node)

Any information would be appreciated.