Stripe size is typically 64K, which means that the data is split across drives in 64K chunks. It could be 32K or 16K or 128K and so on; minor efficiencies can be had by choosing for specific types of access.
RAID-0 striping and RAID-5 striping with parity offer much higher performance than with a single drive under the right conditions.
Peak RAID performance is realized only when the I/O transfer size is substantially larger than stripe size, which allows the multiple drives in a RAID to operate in parallel (simultaneously). This is where the speed gains of RAID striping variants come from.
For example, with a stripe size of 64K, reading 256K of data allows 4 drives to each read 64K in parallel (assuming best-case alignment at 256K boundaries, but this is often not the case). Doubling that to 512K ensures higher efficiency, but even with 1MB transfers, peak performance is still only about 65% of peak peformance for RAID-0, as the graph shows.
Since many programs read/write in relatively small chunks, the full performance potential of RAID is often unrealized for real-world applications.
As a user, there is generally nothing you can do about this; it’s the application making the decision. For example, Photoshop uses ~ 1MB chunks for reading and writing its scratch disk, an anachronism that Adobe really ought to fix: observe on the graph below that 1MB I/O size yields much less than the peak performance. It is why a single very fast SSD can perform so well even relative to a RAID, see Thunderbay Mini with SSDs as a Photoshop Scratch Disk.
Shown below are two graphs, one for RAID-0 and one for RAID-5. Click through for article and then click the graphs for a much larger view.