Time between work units: I'm curious about the time it takes my computer's GPU to finish a work unit. I've noticed on my main computer that it seems to stop processing at 99%. It then does nothing for a noticeable length of time (the GPU fans go quiet as it's no longer under load). That final 1% takes a long time when surely the system could be doing more. The advice for Einstein@home it to only run one work unit at a time on a graphics card (mainly due to GRAM constraints), but surely if it ran two, it could at least be getting on with one while it's finishing another...
Limit the number of CPU cores: If a computer only crunches on the CPU and is not used by anything else or at risk of overheating, I will likely have CPU usage at 100%. One also needs to consider if the CPU is Hyperthreading enabled since these are classed as CPU cores also. This can get confusing with regards to setting a percentage; ideally you want maximum processing but not at the risk of bottle-necking, which is more of a risk when using a dedicated GPU.
Suspend when... On a computer like 'Phenom' which has 6 CPU cores and no GPU, I have it run at 100% but it is set to Suspend work while the computer is in use. On computers with a GPU I might use the option to suspend while a particular application is in use that is calling on the demands of the GPU also. Different types of data being crunched by BOINC might have a different impact. Instead, I may just manually disable GPU crunching or all crunching temporarily should I notice an sluggishness in what I'm trying to do.
Limit Network Usage: During normal usage I limit network usage to 100 KB/s. I only remove this cap when first setting up a project on BOINC so it can more quickly download all of the initial files and work units. I cap my usage because my broadband connection is not very fast and it will impact me if one of my computers is suddenly downloading a bunch of work units.
Storing work units: I set my
computers to download 2 days
worth of work, plus an
additional 2 days. This was
recommended for the SETI@home
project to help keep stocked up
during server outages, while not
being overly demanding on
servers when they come back
online (when lots of users will
all be demanding work).