The clock recovery component looks for the portion of the signals with most correlation, but in some cases when the signal undergoes distortion due to propagation in a fiber the two signals are much different not only from delay. In these cases the clock recovery component will not work.
In your second case the result is a bit more surprising. Try adding a small optical delay, the clock recovery should find this delay no problem. If the signal is exactly the duplicate, which hopefully is quite rare, then there might be some type of numerical error that it does not find the correct 0 time delay.
Regards