Why isn't Azure Load Balancer using round-robin?

If I use the Azure Load Balancer in front of a set of Web Roles then the default behavior is that a hashing mechanism is used to determine which machine an incoming HTTP request should be routed to. This hashing mechanism takes source IP, source port, destination IP, destination port, protocol type into account.

This means that regardless of number of web servers in my cluster, a single connecting client  can place very high load on a specific web server, making the web application unusable for all other clients which ends up on that specific server. In other words, the load balancer does not attempt to distribute actual requests in a balanced way - only the clients connecting.

We saw this issue with one of our clusters today, where a specific web server was under very heavy load, while the other servers in the cluster was idling. We've confirmed using New Relic and server log analysis that this issue was caused by unbalanced loading of incoming HTTP requests.

Why have this method of distributing load been chosen over a plain round-robin where each server can be more or less guaranteed equal number of incoming requests? Is there  a way to configure the load balancer to use round-robin instead?

---

Nitramafve