Hi,
From the documentation of removedelay I can see that the prediction on Nets in question will result in output of one-time step ahead. It is clearly mentioned that for original network if the output is y(t) then removed delay network will return y(t+1). So, both predictions will be same but in the removed delay network one extra prediction will be there because the time delay is not present. You may try the example given in the documentation to confirm the same.