I think the reasoning in the experiment is a little flawed. Given that each task is so small, the cost of a context switch in the example is significant.

Most people multi-task because they are waiting for the first task to finish/ get rid of the monotony of doing the same thing over and over again etc. in which case multi-tasking can significantly improve output. Its kind of like the OS running multiple processes on a single core. If you are waiting for something, you don't need to hold up everything else..

I think the biggest issue is the set-up when jumping from task to task. I find if I leave a project for too long, it takes me several minutes to figure out where I am so in the process.

Set-up (SU) and set-down (SD) is the issue - breaking concentration. SU(Task#1)SD; SU(Task#2)SD; back to SU(Task #1)SD; SU(Task #3); SU(Task#1). Too many SU and Too many SD per task easily doubles the total time for all three tasks.