The problem with RAID5 is that if one of the hard disk drives fails in the exact moment that a data recovery operation is being conducted, the system fails and there is data loss, i.e., the system isn’t capable of recovering the data that started the data recovery operation and also data is lost because one of the hard disk drives is defective.
 
While for small RAID systems the probability of such scenario to happen is very little, for big RAID systems such situation isn’t so remote. During IDF Spring 2005 Intel provided a study pictured in Figure 1. It shows the probability of failure on a RAID5 system for three different configurations:

• Scenario 1: RAID5 system with five enterprise-class 30 GB hard disk drives (120 GB total). Frequency on which data recovery operations is necessary: one every 23 years (orange on the chart). Probability of a system failure during the data recovery operation: 0.12% (i.e., one error every 834 data recovery operations, in green on the chart).
• Scenario 2: RAID5 system with five desktop-class 300 GB hard disk drives (1.2 TB total). Frequency on which data recovery operations is necessary: one every 2.3 years (orange on the chart). Probability of a system failure during the data recovery operation: 11% (i.e., one error every nine data recovery operations, in green on the chart).
• Scenario 3: RAID5 system with 50 desktop-class 300 GB hard disk drives (15 TB total). Frequency on which data recovery operations is necessary: one every 3 months (orange on the chart). Probability of a system failure during the data recovery operation: 70% (i.e., one error every two data recovery operations, in green on the chart).

click to enlarge
Figure 1: Data loss probability on RAID5 systems.

As you can see, the probability of data loss during a data recovery operation increases with both the number of hard disk drives in the system and the capacity of each hard disk drive.