Why don’t the System and User measures of CPU usage give the same number? Surely they’re measuring the same thing?

Most operating systems give measures of CPU (and other resource) usage from the point of view of the system as a whole, and also from the point of view of the workload that runs on it. Since the "workload" is what causes the "system" to be busy, you would expect them to be the same. However, that's not the whole story.

Firstly, there are always some system activities that don't really belong to any workload. A good example of this is scheduling - the process that the operating system carries out to determine which of several candidate tasks to run next. Other system costs that cannot be charged to any workload include interrupt handling. Although an I/O may have been carried out for a particular user task, the operating system has to carry out some low-level processing of the I/O interrupt before it can determine its destination.

Secondly, particularly in UNIX systems, there is a problem with the level of detail that can be recorded for user tasks. Most UNIX implementations account for user CPU time in multiples of one hundredth of a second. Any task which uses less than that amount of CPU time may not get "noticed". A hundredth of a second may not sound like much, but if these very short tasks run millions of times per day there can be a very large shortfall indeed.

The bottom line is: trust the system measurements rather than the "user" or "accounting" measurements.