We need to start a thread to report WU size estimates for the various projects.
I will put together my experience shortly in an HTML file and post it here.
If anyone else has some misc. WU size experience (in length of time preferrably) please post it here along with the machine specs that it ran on.
Iām sure there are some websites out there that already provide this info - please feel free to post URLs to them as well 
This is a short list from the last 24hrs running these projects.
UFluids
Intel Core 2 Duo 6400 @ 2.2Ghz 1GB RAM
Time to process workunits average 16.08 mins.
AMD Athlon XP2700+ @ 2.15Ghz (Barton core) 512MB RAM
Time to process workunits average 18.06 mins
AMD_64 X2 4200+ @ 2.2Ghz 1GB RAM
Time to process workunits average 38.34 mins
(note: This machine is running 2 instances of UFluids and 1 instance of EON)
Proteins@Home
AMD_64 X2 4200+ @ 2.2Ghz 1GB RAM
Time to process workunits averages 33 mins.
AMD_64 X2 4200+ @ 2.2Ghz 2GB RAM
Time to process workunits averages 49 mins.
(note: This machine is running 2 instances of Proteins@Home and 1 instance of EON)
Rosetta@Home
AMD_64 3000+ @ 2.0Ghz 512MB RAM
Time to process workunits average 2hrs 48mins
QMC@Home
AMD Athlon XP 2800+ @ 2.08Ghz (Barton core) 512MB RAM
Time to process workunits average 9hrs 42mins
Riesel Sieve Boinc (LLR)
Intel P4 3.0Ghz @ 3.0Ghz 1GB RAM
Time to process workunits average 14.51mins
PrimeGrid (PrimeGen)
AMD Sempron 3100+ @ 1.8Ghz 512MB RAM
Time to process workunits average 13.29mins
(note: This machine is running 1 instances of PrimeGrid and 1 instance of EON)
AMD_64 X2 3800+ @ 2.0Ghz 1GB RAM
Time to process workunits average 13.14mins
(note: This machine is running 2 instances of PrimeGrid and 1 instance of EON)
As you can see from the differences in run times EON when ran as a second project affects each project differently, not sure if this is a memory bandwidth issue? It appears to have little to no effect on PrimeGrid, but slows UFluids and Proteins@Home down considerably. Further testing is called for to determine which projects will run best with EON.
One other thing, several projects have a number of quasi sub-projects/clients running under one banner. PrimeGrid has "PrimeGen" the orginal project and "TwinPrime Search" which runs the LLR workunits. The workunit run times are different. Same thing goes for Riesel Sieve project, they have the orginal BOINC sieve and BOINC LLR plus their manual sieve and manual LLR clients. There maybe others?