The number of cache misses will be the same backwards and forwards. Each time you address outside a cache line, the cache will fill an entire line from DRAM. Whether you fill waiting on the last or the first byte doesn't matter if you have a smart cache/memory controller that fills the desired bytes first (I know PPC used to do this; haven't actually checked for modern CPUs).

So, array traversal forwards or backwards are pretty much the same from a caching point of view -- and significantly better than linked lists, where you're guaranteed a cache miss for EACH element you visit!