In cache-based computers (as opposed to architectures that directly work memory-to-memory such as the Cyber 205) data elements first need to be brought into cache before they can be brought into a register in order to be operated on. Most of the times, the architecture is such that not a single element, but rather a few elements, collectively called a cache line, are moved from main memory to cache.
The existence of cache lines implies that programs that observe locality of reference can achieve a high performance.