I agree with the arguments given for the OnBoard implementation likewise costs are negleglible compared by the Stratix V device or having interfaces to communicate via backplane etc.
Nevertheless there exist applications that are cost driven using Cyclone IV devices ans e.g. communication via CAN-Bus (e.g. motor controls installed at the drives). For this the relation between the additional costs for material (IC,..) population and test of the USB interface (which is still useless in industrial world for communication) and the FPGA is no more something you can ignore.
Maybe the majority of FPGAs are used in applications with the USB interface either being already required or adding additional value to the system / costs are no issue, but there are still other designs a direct implementation is really no add-on value (especially as even a not used USB interface degrades the MTBF as the chips are still connected to power supply, ...).
(At least I hope its not only me using FPGA in some cost sensitive designs).
Best regards