Tests that use replSetSyncFrom can fail on slow machines. They can fail because immediately after a test issues a replSetSyncFrom command, the "sync source is 30 seconds behind another viable source" logic can kick in and change the sync source again.
This logic is based on optimes gleaned from heartbeats, so if there is a period of time when a server takes longer than 30 seconds to replicate a write (for example, at startup time when servers are preallocating files on mmap), the test is in danger of hitting this situation.
Using a failpoint to disable this logic would prevent these types of failures.