Techniques for processing events in parallel are provided. Multiple publishers publish events in parallel to a plurality of channels. Each channel is a queue that maintains a list of events that are awaiting to be processed by one or more slave processes. In one approach, all events that need to be persistent are published to a persistent channel, whereas all events that do not need to be persistent are published to one or more in-memory channels. In another approach, for each event, a publisher determines which channel of the plurality of channel to publish the event. The determination is based, at least in part, on the number of events in each channel and, optionally, the type of each event in each channel. In either approach, each event from a single publisher is published to the same channel.