Dear Eric,
These are very interesting questions I am concerned about as well, and for which I don't have so many answers. However, here are some general thoughts:
1) I am afraid it is difficult to isolate sharing from many potential confounds responsible for RUTF/RUSF program reduced effectiveness (i.e., acceptability, breastmilk displacement, adherence to administration messages like proper hygiene practices, which food to avoid when giving the product, storage conditions, etc.). Most authors I read rather formulated the problem as facilitators and barriers to home-based supplemental/therapeutic feeding, or simply described the use and acceptability of such products. Then, they used mixed methods to
--> first, point out these facilitators and barriers (qualitative), and second,
--> try to measure to which extent each of them exist (quantitative) or decide to focus on some of them.
Also, they tried to circumscribe their research to intra-household versus a broader dimension of sharing depending on their needs. Depending on what your colleague observes in Senegal, he/she can think about what suits best to the context.
2) Proportions: again, it may be very tricky to directly link the whole 10, 20 or 30% of ineffectively treated children to the unique presence of sharing practices because of the counfounds I was talking about. Hence, in my opinion, the question of which threshold to use as an alert unfortunately becomes quite challenging (to not say impossible) to address. As an alternative, UNHCR uses the benchmark of less than 25% of households reporting sharing as a program quality indicator within their M&E framework (see p89 of this document http://www.unhcr.org/4f1fc3de9.pdf). Of course, questionnaires must be designed properly and interviewers be well-trained in order to avoid respondent bias as much as possible. Also, I don't know how big are your programs, but I would also consider reporting counts. Indeed, 10% of 100 beneficiaries is 10 persons, but 10% of 10000 are 1000. Presenting both figures can help visualise the extent of the problem better and alert stakeholders in a proper manner when needed.
You may be interested in reading two works done in 2011 (http://www.ncbi.nlm.nih.gov/pubmed/22867910) and 2015 (http://www.tandfonline.com/doi/abs/10.1080/19320248.2014.962772?journalCode=when20) on this topic. I have other works compiled as I read a lot on the topic a few years ago so feel free to write me ( aurore.virayie@gmail.com) if you need further documentation.
All the best and please keep us posted