Went to near to get some insight (pun intended) into why Rsw.in and Rsw.global.in don't agree nearly as well as in SOAS. Didn't solve this problem, but did find that the shield on Rlw.in was cocked -- perhaps even in the field of view of the sensor -- I took a photo. Tom thinks he may have loosened it when he last changed the dessicant (more than a week ago), though he didn't notice it being off when he cleaned 2 days ago. The logical conclusion is that it was loose and recent winds skewed it more (though winds haven't been <that> high in the last 2 days...)
Also decided to clean the radiometers while I was there, since I could imagine a film of dust on the SPN1, though the tissue didn't come away dirty. There were spider webs around. At most, the change of either Rsw.in or Rsw.global was 3 W/m2 -- not enough to explain the large differences we've seen.
Cleaning done about 15:30, though I was standing near the radiometers starting about 15:15. 