Load balance based on Host CPU/RAM utilization?

You can use Observed load balance or Predictive load balance but that is based on a derivative (response characteristics) of the pool member that are somewhat based on cpu/mem/io.

However, to check the ESXi hosts themselves? Not easily considering that they are not in the member pools themselves. You would have to create an external monitor script to do the dirty deed on each ESXi host.

I generally am loathe to ever use an external script in the LTM, but I don't see a better way to do it