This is a question straight out of the domain of scientometrics - a field I have had some limited experience in. There is no one source which gives you your answer; rather this is a contextual, complicated and rather nuanced question which you have asked. The true answer is it depends on the field. For instance in the field of ICTD (Information and Communication Technologies for Development), the proportion of articles is very high (which makes intuitive sense) but this may not be true for other fields. Remember also that the term developing countries has a very shifting definition and is therefore temporally bound.
With this exposition, here are some studies which I can point you towards which have looked at answering some, sub-section of your question. I also include studies on conferences (because certain fields are more conference centric) and also broadly scientific productivity and impact among scientists and researchers in developing countries. This is because these are all highly related and in total, might help you.
This study tells us that only 7.9% of articles in the top 21 Library and Information Science journals are from developing countries or eastern european countries.
This study tells us that articles from developing countries tend to cite other articles form other developing countries and thus, creates a vicious cycle of citations.
This study attempted to define a special collaboration metric for articles from developing countries. It recommended usage of this metric to measure productivity and impact from developing countries instead of the usual suspects (h-index etc.)
This study, specific to Chile attempts to find out root causes of low scientific productivity (i.e. publications) from Chilean authors in Chilean universities. (eg. low priority of funding, lack of interest in research, corruption etc.)
This study is a meta-level article into looking at the scientific productivity in developing countries across the world. It presents a socio-historical approach and also attempt to quantify certain key characteristics of developing countries' scientific productivity and reasons behind the low numbers of article.
This study points out that the present measures of bibliometric/scientometric analysis is skewed towards developed countries and as such is not a good indicator of research productivity in developing nations. The authors point out that alternative sources of databases, specifically presented as a case study of Cuba might provide better insights.
I would also request you to take such quantitative studies with a pinch of salt because they do not often paint an accurate or holistic picture of what's actually going on behind the scenes.