It's very technical why the distribution balance can never be even [low supply or bandwidth/storage limitations to mention a few], but if you select your favorite projects, and the 'If there is no work for... sent me something else available', you will get a nice assortment, I do. Yesterday on one dual core device with zero buffer setting, intentionally, had FA@H, MCM1, OET1, FAHV and UGM1. CEP2 I've not selected, in fact, set the allowed disk use to below 2.5GB to prevent them from being sent.

7.35 mcm1 MCM1_0009718_1057_1 08:38:16 (08:34:34) 12/9/2014 10:56:36 AM 12/9/2014 10:57:04 AM 99,29 Reported: OK + 3022763 49.88 MB 87.82 MB
7.19 faah faah939329_ZINC00283516_x2vg5RTKNUC_00_0 05:02:48 (04:54:18) 12/9/2014 6:58:18 AM 12/9/2014 7:02:18 AM 97,19 Reported: OK 3022763 77.35 MB 123.36 MB
7.35 mcm1 MCM1_0009710_9102_0 08:36:23 (08:31:37) 12/9/2014 2:18:21 AM 12/9/2014 2:18:56 AM 99,08 Reported: OK + 3022763 42.03 MB 87.86 MB
7.19 faah faah939273_ZINC00844400_x2vg5RTKNUC_00_0 06:18:44 (06:15:40) 12/9/2014 1:55:13 AM 12/9/2014 1:54:54 AM 99,19 Reported: OK + 3022763 73.67 MB 114.21 MB
7.35 mcm1 MCM1_0009704_4948_1 08:40:29 (08:36:21) 12/8/2014 7:36:21 PM 12/8/2014 7:36:30 PM 99,21 Reported: OK + 3022763 42.69 MB 87.87 MB
7.08 oet1 OET1_0000091_xZAGP_1016_0 05:02:52 (05:01:15) 12/8/2014 5:42:00 PM 12/8/2014 5:40:45 PM 99,47 Reported: OK + 3022763 45.77 MB 44.35 MB
7.32 fahv FAHV_x3OTS-AS_0885265_1987_0 01:43:48 (01:42:48) 12/8/2014 12:39:04 PM 12/8/2014 12:39:23 PM 99,04 Reported: OK + 3022763 38.81 MB 37.32 MB
7.08 oet1 OET1_0000071_xEBGP_1912_0 05:14:21 (05:03:31) 12/8/2014 3:29:33 AM 12/8/2014 3:30:26 AM 96,55 Reported: OK + 3022763 45.94 MB 44.46 MB

Now and then technicians go into the feeder matrix to set how many slots each science gets allocated to re-balance, where in the long run one would get quasi even runtime for the various. If there are many volunteers though who set exclusive sciences like on the low availability UGM1+OET1 that suck their slots allocations empty all the time, there's very little left for those that choose a wider selection, so they would predominantly get MCM1 and FA@H.

In a nutshell.

BTW, this snip sentence reads contradictory "none of my systems have received work units for only one project." Says something like your systems are getting work for multiple rsearches? That will be the case if there's a situation where the particular science slots are empty at the moment of request _And_ have the "If there is no work for...sent me something else available". At any rate, my dual is getting a selection of random work, whatever is available in whichever slot (there's 3000 or so getting refilled based on those feeder matrix weightings). Taking OET1, there's only 1 work unit available of this flavor per 4 seconds, when there are 48 requests for a work unit in that time period. The statistics make it then very low chance to get the low priority fed work units [not to drain the pool in the first 30 minutes of the day].

OET1 is special. It would have been wider circulated, but a substantial technical issue cropped up for the "flex" workunits, of which there is an abundance in the library. We'll have wait until they pass beta testing, no word if alpha is making any progress. Till then only the "rigid", but there's only about 20 thousand a day. That's only 29 minutes work distribution on 1 million per day.
UGM1 will be low priority for all the time, is the indication. The work units are too big in bandwidth and storage. Would they be allowed a bigger quota, it would put the whole grid in jeopardy, a continuous stream of 'Server load too high, backing off 3600 seconds'. Can't have that of course.
CEP2 If you got 2.5GB storage per task and 1GB of RAM, you can opt in, they are in fact highest feeder priority if selected. Trying and trying, there's never more volunteers to do 25-30Kwork iunits a day.
Rob

Manual Update at 12:54 and WUs are flowing!