When you have an application that must log information (for auditing, debugging, etc.) but still run as fast as possible, it is rather wasteful to always dump fully-formatted human-readable trace records. It's far better to dump a short binary record indicating the message identifier, parameters for the message (if any), timestamp, process/thread identifier, etc. that can be processed later for human consumption using a separate "trace formatter" tool. This way you save on processing time and disc space but make it slightly inconvenient to view the log files.
On UNIX-like systems, utmp and wtmp records are created and processed this way. I have also seen this kind of logging in IBM's AIX operating system and its CICS transaction processing monitor. Why then do several modern "high-performance" applications still insist on using the slower and more bloated method?