After a Scheduler has started a multiple instances of a service, it is usually not enough to just provide a way to learn where the new instances are via Service Discovery. In most cases, workloads going towards these services, such as HTTP requests, need to be balanced across all available instances.
Load balancing components provide that functionality, usually by routing requests through themselves in form of a reverse proxy.
Some Service Discovery mechanisms, like Mesos DNS can also be regarded as load balancing as it allows to distribute requests from clients relatively seamlessly.