The settings do not LIMIT how often BOINC Manager can connect. They simply allow it to plan the schedule accordingly. So, if you have a network connection and are out of work for some reason, it SHOULD contact the project and get work. Regardless of the cache size you set up, or when it last contacted the project. But that's apparently part of what's not working properly at the moment. So we're just telling it to try and get several days worth of work (more then you really need), to try and keep CPUs from going idle.

You are assuming work is scarce. Check the homepage, there are more then 3 MILLION tasks for you to get your share of.

The setting will cause BOINC to behave as though it better store up enough work to go for up to 2 days without an internet connection. So, checking less often yields request more work from the server. And checking more often yields... well, in your case, an idle CPU.

Yes, until they (Berkeley) fix the BOINC client, please try 2.0 or 3 days in both entries.

Have you tried setting both the "additional" and "connect every..." to 2 days?

Looks like you are running the 6.4.5 BOINC client. It has had some problems requesting the proper amount of work and maintaining proper runtime estimates (DCF). I'd suggest that until these issues are corrected, you increase the number of days of additional work specified in your network preferences to try and bring down more tasks.

You should have a task per available core. If Einstein can run 2 tasks, this indicates BOINC is configured to allow both cores to be used. So, perhaps one of the tasks is waiting for memory? What do you see for the status on the tasks tab of the advanced view?