This is a really great analysis and makes an important point.  I like the idea of using an MyDD poll to get better numbers (wasn't the suggestion to poll likely caucus-goers in Nevada?).  I don't think the major polling operations are going to do anything much different from what they're doing, because their mandate is to produce results that the general public can understand.  You're right to point out that in this instance what they're doing has a (probably not deliberate) pro-Clinton bias.