WCG packages substantially more 'attempts' into a job for BOINC clients than UD hence the longer run time. Therefore doing 7-9 units is no measure versus 3-4 units as at the end of the day the same amount of result is achieved at equal CPU throttle. Why? It is generally recognised that the average runtime on the remaining UD clients on takes much longer to finish a job than the BOINC clients. Longer jobs also improves the data transmission efficiency from the server to the client and back.

If you have been running UD agent at 60%, that was/is a smooth experience. Opposed the present BOINC throttle is crude and works in multiples of seconds. The 6.2 future BOINC release may have a 1/10th second throttle precision, so 60% or any other one gives more of a lesser awareness that it's running.

Alternately, look at the Start Here forum Threadmaster FAQ. It's not as slick as the WCG UD agent throttle to set up but once running, equally smooth in experience.

NOW the 65000 dollar question: With the UD agent doing a HPF2 job in 2 hours indicating a reasonably fast computer, are you any by chance using a PC with a P4 Intel CPU? If so, change the Processor option in the Device Profile to only use 1. The Hyperthreaded P4 has a 'Virtual' second core and BOINC is interpreting that as if you have 2 CPU's in your computer. You do get a bad experience with 2 jobs run simultaneous on such CPUs

I've had problems with older versions of boinc (5.8.xx and previous) especially on older equipment.

However, if you insure that under preference you have boinc set to unload itself from memory when suspended and you have set the activity preference set to run according to preferences, boinc should not interfere with other processes. However there will probably be a significant delay when returning from idle - at least that has been my experience on machines with 1gig of memory or less and running at less than 2ghz.

I would also insure that the OP is running the latest WCG approved version.