> your results indicate that, for even these simple processors, larger caches are needed to provide reasonable performance

Hmm, I think we're reading the results in the opposite way: a cache can be accessed at the same speed as on chip block memory, and several times faster than our off chip memory. Even with just 32 words of cache we get most of the benefit, which is a factor of 4 or 5 as I recall, and even with 32 words we only use a couple of dozen extra slices. This is very cheap!