The question is why are you trying to do this without using heat and autoscale? The main drawback is that you are solving problem that is already solved by Heat and Ceilometer Alarm.

To be fair you do not have to use Heat to autoscale. You can write entire separated small program that just check how busy your pool of VMs is and send API to create new instance when needed. However, this would be poor's man solution. Heat is working it in the elegant way by creating Ceilometer alarm that you can confugure to how sensitive it should be, it works well, I tested it and it is not that difficult.

go pro, go heat+autoscale