Yes, project selection may be an issue at some point and if the GPUs come flooding into Universe@Home I will be running away as fast as possible :-)
My point about dollars per mag was this: when computing the profitability of something, you have to include the operating costs (energy consumption in our case). Dollars per mag misses a big part of the picture.
Yes, I admit I've been promoting dollars-per-mag mantra recently to create some extra attention (GPU mining always being the hottest topic in crypto, pun intended) and to point out that older GPUs are not necessarily weaker when running BOINC. But there are many scientific algorithms which will never be ported to GPU and this low power approach of yours is the future really, although it may not win many upvotes in "show your powerful rigs" threads :)