David your question has not a straightforward answer.
In case you did not follow I went a few months ago through a lengthy discussion and many tests on the IPS, FLOPS, MIPS, GFLOPS, and the issues to measure the WCG grid crunching power and how it relates in points, watts etc..

The problem here is that the only WCG elements we have at hand are points, runtime, boinc manager benchmark, and wattage measured at the socket.

To make it simple and comparable between us I will use the benchmark that comes with the Boinc Manager.
I get on my fastest machine 12'000 MIPS/core. The manager also states that I have a processor with 12 cores. Now this is not completely true. I have indeed 12 threads or virtual cores and 6 physical cores. For me it is unclear about which type of core we speak here.
I will make the assumption that the benchmark is not multithreaded simply because when it is run the system consumes about 9% CPU capacity. I will consider that the benchmark runs as a unique thread on one physical core. On that base I have a total crunching power of
12'000 X 6 = 72'000 MIPS or 72 GIPS
But when I runn WCG the 12 threads are being effectively used. By having tested the effectiveness of Hyperthreading on the Intel 980X CPU we can say that we can get an improvement between 10 to 15% depending on what we crunch. To remain conservative I will say +10%. This boils down to :
72'000 X 1.10 = 79'200 MIPS or 79.2 GIPS
The machine uses 250 Watts when running at full power 12 threads with WCG. So you get the following ratio of:

79'200 / 250 = 316.8 MIPS/Watt

This value is indicative and is a relative measurement which can be compared across our machines. But it is not the real value. We do not know how many Integer operations or Floating Point operations are used in each WU and from each project. The mix is different between a C4CW WU and a HFCC WU and so on.
If you measured your ratio as I did then my answer is appropriate.