At a previous company, we too would administer a technical test. Our pass rate was close to what was described in the article (40% for ours vs 25%). However, our test was incredibly simple. At most, it should have take a competent developer two hours to complete [including writing comments and a README].

The assignment was to read a file containing a list of numbers (some formatted incorrectly, so there was some very simple parsing logic involved), call an API using each correctly formatted number as a parameter, and store what the API returned to a file. I am to this day stunned that 60% of people who passed a phone screen could not solve this task. Note that we gave them the input file, so it wasn't a matter of an edge case tripping them up or them getting one input file but the test input file having some other edge case.

My point here is that it may be possible to get the same screening value with much less investment from the candidate.