I'm familiar with this testing strategy. It has one major upside: It allows an exam to test the material in great depth without being enormously long.
But it also has one major downside: Students may not learn all of the material equally well. If the random point in the material space that the exam focuses on happens to be a point that the student didn't learn as well, the exam result will appear as if the student learned all of the material poorly.
In other words, the exam has higher random variance.
This happened to me in my computer graphics class. The first exam had a multi-part problem where the first part involved quaternions. I was familiar with quaternions from my time as a physics major, but did not remember the professor introducing them. I reviewed my notes afterwards and found that he did in fact mention them in passing in a lecture two weeks previous to the exam, but did not call particular attention to them nor did he assign any homework involving them.
But it also has one major downside: Students may not learn all of the material equally well. If the random point in the material space that the exam focuses on happens to be a point that the student didn't learn as well, the exam result will appear as if the student learned all of the material poorly.
In other words, the exam has higher random variance.