Load balancing is just a waiting connection to all pools in your pool list. Work is then round robin rotation based on pool work quota, meaning each pool simply takes a turn. But there are also other factors in the work quotas, a disconnect will reset the pools quota, and I believe it adjusts quota based on rejects too.
I used to run load balancing between two servers and it was never 50/50, even over a long period. If you want to measure pools correctly, the load balancing isn't going to work. You'll have to setup a separate cgminer instance for each pool, preferably with identical performing card for each pool.
It's good to hear from someone who has read the code. So, you think the uneven balance word not average out over a test like this and would favor one pool all the time?
Chad is going to dedicate 1 rig per pool, thus avoiding this issue.