You have to have a value set in both settings in the newer version of boinc. FWIW I set the minimum to 0.10 and then set the max to the days of cache I want to keep.
You can control the amount of CEP2 tasks you want to run at the same time by placing an app_config file into the WCG project folder.
Copy and paste the text below into notepad. Rename it app_config.xml and place it into the WCG project folder. If you haven't done so already you'll need to set your OS to show hidden files and folders because the BOINC folder that contains the project folders is hidden by default. You'll need to shut down boinc and restart it for the changes to take effect.

<app_config>
<app>
<name>cep2</name>
<max_concurrent>2</max_concurrent>
</app>
</app_config>

You'll also have to change the amount of CEP2 tasks you can receive at one time in your preferences on the website. The default is 1. If you want to run 2 change it to 3 or higher.

Thanks for your info and suggestions, nanoprobe and branjo.

BOINC 7.0.42 is now working very well for me, running 2 CEP2 WUs and maintaining the work cache quite well ...

I have determined that "Max additional work buffer" is in addition to "Minimum work buffer", not in addition to zero extra work.
In other words "Max additional work buffer" is the width of the window between the minimum and maximum levels of work. The client will start fetching work if it estimates that the level is below "Minimum work buffer", and will stop fetching when the level is above ("Minimum work buffer" + "Max additional work buffer").
I am currently running "Minimum work buffer" = 1.00 and "Max additional work buffer" = 0.10
I got the clue to this behaviour by temporarily setting <work_fetch_debug>1</work_fetch_debug> in my cc_config.xml .
Then, every 60 seconds the BOINC Manager Event Log (formerly the Messages tab) shows that the client is checking the cache level. Here is one of the lines it writes in the Log with the settings above:
> 3/01/2013 6:49:54 PM | | [work_fetch] target work buffer: 86400.00 + 8640.00 sec
This indicates that BOINC is adding my 0.1d "max additional" work to the 1d "minimum" as the test criterion.

Thanks to nanoprobe, I realised that I'd placed my app_config.xml in my main BOINC data directory instead of "main BOINC data directory"/projects/www.worldcommunitygrid.org . Now it's working like it should.

I also found that the limit imposed by <max_concurrent> overrides the run-order priority that WUs that are Waiting to run (WTR) have over ones that are Ready to start (RTS). (Members who micromanage the order of execution of WUs will know that BOINC runs any WUs that are WTR before any that are RTS, high priority panics excepted. Forcing a WU that is RTS to run for a short time, then suspending and resuming it, will leave it as WTR, and this is the main way that run order can be manipulated by the user).
So what does this finding about <max_concurrent> mean? This parameter limits the max no of WUs of a type to be run simultaneously, but it does not enforce a minimum. If 1 or 2 of my 2 CEP2 WUs finishes and the next RTS WU in the cache is not CEP2, and there are no WUs that are WTR, BOINC will run the next (non-CEP2) WU that is RTS, while I want the threads that were running CEP2 to hand over to fresh CEP2 WUs further down the queue. I can make this happen by setting some or all CEP2 WUs in the queue to WTR.
If I micromanage in this moderately easy way, I can ensure that 2 and only 2 CEP2 WUs are running at all times, without having a very large value for "Number of workunits per host for The Clean Energy Project - Phase 2?" in my device profile.

@branjo: I don't see how network activity times are relevant to this topic. My BOINC clients don't have to explicitly connect/disconnect from the Net because I have an ADSL landline connection that is "permanently" connected. However, when I'm away from home for more than 1-2 days, I set "Network usage allowed between 08:45 and 08:30 every day", plus Activity menu > "Network activity based on preferences", and power off the modem/router for 1 min at about 08:38 using a mains timer, to avoid running out of work due to a connection hangup. (which happened a couple of times before I set up this scheme).

You wrote that BOINC Manager "would not download new work without manual "Update"s." The changes in Network Tab I have posted above have solved this problem for me



In version 7.0.42 both app_info and app_config get read only on BM startup. As far as I remember, SekeRob have posted that developers are working on the option you mentioned via "Read config file".

Cheers

Edit to add: Thanks for thorough feedback

I am sorry for misinformation . I have forgotten that I have used app_info with 7.0.36 and after upgrade to 7.0.42 I have completely replaced it with app_config.

Cheers