The impact of different computer architecture models

In comparing a Harvard design with Von Neumann, the Harvard design should be more performant: if you have an instruction which requires an item of data to be fetched from RAM, on a Von Neumann design, this will require two cycles (one to fetch the instruction, and then one to fetch the data; only one item can be transferred on the bus at a time).

In a Harvard architecture design, as there are separate buses for data and instructions, both the instruction and the data can be fetched at the same time. This allows the operation to be completed in a single cycle.

Generally, Harvard will perform better than Von Neumann; for this reason, even though modern CPUs present as Von Neumann, they often use a Harvard-style set of data and instructions buses internally to connect cache memory to the CPU.