The problem is not bandwidth, it's that on a single IDE channel, only one device can be communicating at a time. For RAID0, this means that the chipset has to send data to one drive, wait for whatever acknowledgements, then send the stripe of data to the other drive, wait for the ack, back and forth. It's still faster than just a single drive, presumably, because the controller doesn't actually have to wait for the mechanical write of the data, but it's not the simultaneous read/write of independent channels.
There's no technical reason for it not to work, it's just not an optimal setup. With RAID1 it would actually probably be slower performing, since you're in essence having to wait the same amount of time for each drive to be written to as you would for a single drive, so you get twice the latency. Reads would still be only as fast as a single drive probably.