I can see how a datastream/channel can be used to send commands to a device (e.g. an actuator). The device can periodically poll the channel for incoming commands, but if the device has no storage of its own how can it tell which commands it has already received/processed?
This all depends on your implementation and hardware choices. And the real answer to this question lays far beyond the scope of Xively. You say that the device has no storage of its own, but I assume it has some kind of volatile memory at the very least.
The best thing to do would be to store the timestamp of the last datapoint that was received and compare it to whatever current data you have. If the timestamp is greater than the one in memory then you know it is new data.
An alternative to HTTP polling would be to use a socket with some kind of publish/subscribe interface that will allow you to received only new data from the server. Xively offers this on it's TCP, WebSockets, and MQTT socket servers.