Before I go file a new issue, I at least want to comment here to see what you think Hoss Man and of course anyone else who may notice ( Christine Poerschke touched this stuff too I think?). I claim that "useCompoundFile" in solrconfig.xml (directly within <indexConfig>) is named in a way that is misleading, and Solr chooses a default that is not ideal.
<useCompoundFile> is only about flushed segments from newly indexed documents, yet its setting name doesn't reflect it's limited to just those new segments. We don't even say this in comments or the ref guide (I'll happily fix this). I can understand how this came to be, it's mimicking a like-setting in Lucene IndexWriter which arguably is also misleading for the same reason. But I don't think we need to mimick the name Lucene chose. I could propose alternative names but I just want to throw this statement out there to see if it resonates or not.
Secondly, note that Solr defaults to useCompoundFile=false; Lucene does true. Given a merge policy with a noCFSRatio of > 0 (by default we use TieredMergePolicy which uses 0.1), merged segments with fewer than 10% of the index will become CFS. This is a schizophrenic configuration! Segments start off non-CFS, then get merged to CFS (assuming it's still small enough), and then eventually become non-CFS again. It flipped back then forth. That's weird. Instead, I propose that only if noCFSRatio is 0, then default useCompoundFile to false since it'll never flip-flop. Otherwise, default useCompoundFile=true (which given default MP settings, this will generally be the case).