I played with this device at one point. Both things are true.
There is some LED current threshold which turns the output on and another (lower) which turns it off. So if your current waveform isn't symmetric around those points, output duty cycle has no reason to be 50%.
At high speeds propagation delays become significant and they may be asymmetric and dependent on things like output pullup resistance and input current limiting resistor. Getting 50% duty cycle at 1Mbps takes some tweaking.
In fact, propagation delay is so long that it approaches one full bit time at circa 1.3Mbps, right before the device gives up and starts dropping bits. It's fun to watch on the scope as input and output change their state almost in unison, even though it is an inverting device.