“Checks Returning I/O” reports the number of times that a requested I/O had completed when an engine checked for disk I/O.
For example, if an engine checks for expected I/O 100,000 times, this average indicates the percentage of time that there actually was I/O pending. If, of those 100,000 checks, I/O was pending 10,000 times, then 10% of the checks were effective, and the other 90% were overhead.
However, you should also check the average number of I/Os returned per check and how busy the engines were during the sample interval. If the sample includes idle time, or the I/O traffic is “bursty,” it is possible that during a high percentage of the checks were returning I/O during the busy period.
If the results in this category seem low or high, you can configure i/o polling process count to increase or decrease the frequency of the checks.
See the System Administration Guide.