Maybe I can explain, I've seen the same issue with Kprobe and my Liteon. It's basically a resolution issue....with Kprobe you can delete the highest value spike. Some spikes appear to be single lines but are actually multiple, consecutive spikes that require 2 (or more) deletions to remove. The graph just doesn't have the resolution to represent both spikes. Even if it did, your eyes probably couldn't at that size.
This will be even more evident with CDSpeed, since the graphs are physically smaller than Kprobes. I think if you watch a scan, you'll only see a spike drawn when the error count is going up....many times the count will go up more than the spike....this is actually 2 or more lines graphically "overlayed".
These graphs are typically not used for counting total errors. They are used to represent consecutive, un-correctable areas on the disk. This helps to see problem areas on that disk. A 40 count of total PI errors sounds like a good burn, but if they are 2 - 20 consecutive spikes, then you'll have a nice skip on that disk at the location of those spikes (which may look like one spike on the graph).