Today my laptop downloaded three tasks instead of one. I have my, Boinc - Store at least n days of work set very small
intentionally. When I let it run Milyway, I want it to finish and then it goes back to regular life.
Today, when the task downloaded, the estimated time was very short, so it downloaded more.
After crunching a couple of minutes, a realistic estimated time appeared.

In the bigger picture, these might time out or at least you're less likely to get things back in the order they were sent.
In an even bigger picture, if this is happening to a lot of people, you'll have a lot of timeouts, but I'll admit I probably don't run like the typical crowd.

The wrong GFLOPS estimates, which are causing wrong run time estimates, are a known issue, however no fix for it so far. Yes, you are not running BOINC like "the typical crowd", not sure why your laptop needs to finish the work before "it goes back to regular life" since Milkyway uses very small amount of RAM, so it should be possible to use it while Milkyway is running.

To speed up the correction of the estimated run time use this as app_config.xml for Milkyway:

<app_config>
 <app>
  <name>milkyway_nbody</name>
  <fraction_done_exact/>
 </app>
 <app>
  <name>milkyway_nbody_orbit_fitting</name>
  <fraction_done_exact/>
 </app>
</app_config>

Works only for running tasks, but since you have small cache, it should help anyway. If Milkyway is the only project you run, you can also set it's ressource share to 0 and than BOINC will only request as many work units as it needs to use all cores. So you will have no cache at all, BOINC will request a new task around 30 seconds before it finishes any of the running tasks.

---

I have been running a project for almost a month now. My assumption so far, is that my project will never be completed, it just inserts new tasks. Correct assumption?