The problem for me is not technical per se, but that it adds another element of complexity to an already complex programming environment and has implications for human resourcing.

Becoming expert in a new language goes beyond learning syntax; one must also learn code conventions/style, compiler/transpiler quirks, runtime quirks, and library/framework ecosystem. In CoffeeScript the last two may be moot, but the first two remain. Additionally, to be an expert in a transpiled language, one must be an expert in the target language. It is much harder to find someone that has this knowledge and experience in both CoffeeScript and JavaScript.

This begs the question of each web dev team: if you are already experts in JavaScript, is the value-add from the transpiled language worth the cost of increased solution complexity/cognitive load, and smaller talent pool? My take is that CoffeeScript doesn't meet that bar for most teams.

For individual projects, or isolated one-off tools, these non-technical issues are moot. Also, server-side may be less of an issue since complexity is reduced by not having to write for multiple runtimes (only Node.js and usually without DOM).