When you see a latch free wait event in the V$SESSION_WAIT view, it means the process failed to obtain the latch in the willing-to-wait mode after spinning _SPIN_COUNT times and went to sleep. When processes compete heavily for latches, they will also consume more CPU resources because of spinning. The result is a higher response time. In this case, the high CPU utilization is a secondary symptom. This false CPU utilization level can falsely impress your capacity planner that the server needs faster or more CPUs.
The TOTAL_WAITS statistic of the latch free wait event in the V$SYSTEM_EVENT view tracks the number of times processes fail to get a latch in the willing-to-wait mode. The SLEEPS statistic of a particular latch in the V$LATCH view tracks the number of times processes sleep on that latch. Because a process has nothing else to do but sleep when it fails to get the latch after spinning for _SPIN_COUNT times, the TOTAL_WAITS should equal the sum of SLEEPS as shown in the following code. However, there are times where TOTAL_WAITS is greater than the sum of SLEEPS. This is because SLEEPS statistics are only updated when the latch GET operation succeeds, and not on each attempt.
Because latch free waits are typically quite short, it is possible to see a large number of waits (TOTAL_WAITS) that only
account for a small amount of time. You should only be concerned when the TIME_WAITED is significant.