A LAG/AE does not load balance. Only way to actually load balance is to send one packet over LAG link #1, then next packet over LAG link #2, and so forth. This leads to potential for out of sequence packets, so very few vendors (like ALL) do not implement LAG/AE hash this way. With lots of flows, the probablity of the hash creates basic load balancing. This is why on more advanced (generally more costly) products hash algorthm maybe selectable or programmable.
Not sure why Cisco states what you wrote, but Juniper does not claim to load balance, instead the term is load share.