Depends on your view of subsistence, i.e. strictly speaking the proportion could be considered 0 if you view subsistence as supporting all nutritional requirements by home production. Using an extended definition, some output may be sold to cover extra basic household expenditures and hence that small percentage of farm output should be compared to basic consumption levels. Empirically, I would try to let the data speak for itself by visually inspecting the distribution or looking at the impact of the cut-off point chosen on your final results obtained.

Thank's a lot Yann. I fully share your opinion, however a new question arises in this context: shall this threshold be identical for all countries or context specific. I plan its using in an international farm typology for comparing farm types between different countries. Will it ensure comparability if it will be context specific, i.e. will differ from country to country?

I guess the best option would be having a unified definition across all countries. Note that when defining the cutoff point in relative terms you still account for country-specific levels of output. (See also how the EU classifies farms into typologies based on Standard Output (SO): http://ec.europa.eu/agriculture/rica/methodology2_en.cfm)