Boy you've really summed it all up right there! Let me see if I can clarify some points...

correction: if you set to 1.5 days, BOINC will try to keep 36 hours of work on-hand, not 18 hours.

Credit Hogs: Some projects run low on work units. So people that want to participate cannot when the project runs out of work to send. The main objection is to people that download a huge pile of work, and then are unable to process it. It doesn't sound like you fit the profile.

Setting frequency to connect: yes, this setting is intended to give you a control on the size of your backlog of work. If you had fulltime network access, and were running several projects, it wouldn't matter if one project went down for a few hours, you'd still be happily crunching away on work from the other projects. And so you could keep your machine fully utilized. If you only do work for one project, and that project has an outage or runs out of work, you may be sitting idle for a time, unless you've got work in your backlog. So, that's why the recommendation changes a bit when you only do work from a single project. If you have 2 days of work on-hand, then the project could be down for a day and you would still be fully utilized doing work.

Timeframe of work units: Yes, each work unit comes with a due date ("report deadline" shown in the work tab of BOINC manager). If that piece of work went to 3 users, and 2 reported back the finished results and the third has not reported by the due date, then it will be sent to another user to bring in that third completed result. But it resets the clock on when you can expect your credit for that work unit. And if that user keeps a huge backlog, it might take their machine several day to begin working on it, thus holding up credits for the 2 that already completed it.

Limited network access - dial up: yes, if you are dial-up user, you'd probably want to maintain a larger backlog (i.e. define a larger number of days on the connect to network every... setting). That would give you some slack on how often you need to connect to keep your machine busy, and report back your results in time.

Changing the "connect to network about every N days" setting: The MAIN LESSON is to make changes to this value very gradually. I mean when I started, I was a dial-up user and set that baby straight to 10 days. I got a pile of work for SETI, and then added Climate too, thus splitting my computer time in a way that BOINC didn't have any way to see coming when it had requested all that SETI work.

So, if you change the project "resource shares" or the number of days between connections, you have to keep in mind that BOINC is trying to make estimates and predictions based on the information available at the moment. If WCG is your only project and you say to get 10 days of work... and it does, and then you add another project and cut the resource chare in half... now you've got 20 days of work! Whoops!

Also keep in mind that these settings only take effect once your client updates with the project.

As you say, setting the preference too high doesn't "break" anything. The systems are all set up to be flexible. It just increases the likelihood of you missing some deadlines, and therefore the crunch time on that work unit was essentially wasted. It also may delay the issuance of credits to others that are working on the same "result" (result is what gets broken out into the redundant "work units"), and this is why the people that covet the credits get upset. Someone with a .1 number of days is basically immediately working on the one work unit they download, and will report back results by the end of the day. But they need to understand that not everyone wants to, or is able to, run their machine that way.

Yes, the project has a balancing act between getting more "results" in progress, and getting all of the validation they need. But this balancing act is all based around the deadlines. They basically assume everyone will return results before the deadline, and then if/when they don't then reassign the task to another participant.

Ideally (my opinion), you would set things up in this order:
1) keep your number of days to connect at the .1 default,
2) attach to all of the projects you intend to,
3) get your resource shares balanced they way you'd like,
4) and THEN start tinkering with the number of days between network connections.

By doing it in this order, you avoid pulling down more work than you will be capable of completing before the deadline.

Hope that helps your understanding of how the system works, and your role in it.

I would think the best strategy is to just get the minimum amount of work you would need to span any period of time during which you are not connected to the Internet. If you are continuously connected, then just one or two work units are plenty.

This is because the grid cannot tell if you are still working on a given work unit or have uninstalled the agent. As a work unit ages and if results are not coming back, more agents will eventually be assigned the same work unit so we can get a consensus on the answer. So if a work unit happens to be late in several agent's queues, after some number of days it can look like nobody is working on that work unit anymore and more assignments may be sent out, thus creating inefficiency. This applies equally to the use of UDmonitor.

This is getting rather interesting. I am developing a much better appreciation of why you want to change this value slowly, even if you only run one project on one device, and most particularly if you increase the value. It looks like the exact results one could see will vary depending on your circumstances. If you increase the value quickly, like from 0.1 to 3, you will basically get three days worth of work at about the same time. In other words, I suspect that the WU's you will get in that case are nowhere near as spread out amongst the "pool" of WU's as if you got them one at a time. That could mean that you could end up delaying the quorum and end up getting WU's sent out for more processing unnecessary if you tend to complete and return the WU's slowly. If you are able to return them quickly, you could well end up waiting for others to process them and complete the quorum. That means you could see your points growing a good bit more erratically.

Off hand, I can't think of a problem with dropping this value quickly but that certainly doesn't mean that there aren't any. It would seem that you want to avoid frequent changes and large changes to this value. What is frequent or large will depend on your specific circumstances too.

Also, a subtle point to pay attention to as well, determining how many WU's will be sent to you appears to be based on the estimated completion time (the "To Completion" column on BOINC Manager's Work tab) but this value typically seems to be an upper bound or high end estimate so X amount of work won't necessarily take X amount of time.