As a reader I personally prefer if they do a more complete set of citations, instead of making me follow up a multi-step chain to dig them up, as if I'm a compiler resolving transitive dependencies. I like little history-map sentences like: "This technique was introduced by Foo (1988) and recast in the modern computational formalism by Bar (2009); the present work uses an optimized variant (Bar 2012)."
You could just cite the last paper here, which is the only one used directly, and which presumably itself cites the earlier papers. But it's more useful to me if you include the version of the sentence that cites all three and briefly explains their relationship.
Often half or (much) more of the value of a paper is in the references, and that's not a bad thing. Sometimes it is the first thing I read.
There's no ink shortage, no link limit on the Internet, and every paper has an abstract for quick filtering. As a curious person I want everything that serves to establish the argument cited so I can be guided to papers of interest and get a better idea of where an idea fits in the broader field.
You could just cite the last paper here, which is the only one used directly, and which presumably itself cites the earlier papers. But it's more useful to me if you include the version of the sentence that cites all three and briefly explains their relationship.