Wouldn't choosing stripe size based on your filesystem block size be best?
For example, for our arrays that have 8 drives in RAID10 (4 stripes). We format our VMFS with 2MB block size. Divide block size by # of stripes = 512K stripe size.
Each disk gets accessed exactly once per block read, provided your partitions are aligned. Seems to me like that would provide most efficient access.
Tell me if I'm wrong, but it's working well for us. 140 guests across 5 SAN boxes, 12 LUNs in all.