The multiply need a cycle?

I was able to see the images by manually extracting the links. They are simply unrelated. The first image shows a multiplier with additional input and output registers, that's not the standard implementation of LPM_MULT. To my opinion, the MegaWizard dialog and explanation is very clear in this respect. The second image shows the dataflow of an unregistered multiplier, however ignoring the logic cell delays.

At kaz mentioned, it may be meaningful to add registers or pipeline stages in some situations. It depends on the design speed, of course. You can forget about it first and remember the option, if the timing analysis shows problems.