There are many reasons this “result” is invalid, but the main one is that these tests of complexity are skewed by archaic words that were used (surprise!) more frequently as one goes back in time.
There is also the fact that the circumlocutory style used up until about the early 1960s in formal writing and speech is profoundly unnatural and actually more difficult to understand than phrasings without a hundred barely-connected dependent clauses. (This style of writing is now almost exclusively found in academic papers where it is used as a signaling mechanism.)
Speech and writing complexity does not equal conveyed information or intelligence. Most often, it’s deliberately obfuscatory.
Try reading Richard Feynman write about something extremely complicated and see how clear it can be, and then try reading an average paper about the very same topic to see what I mean.
As another for instance, Shakespeare is only difficult to understand for modern readers due to the archaic and obsolete words, not due to textual complexity. I always laugh when people cite Shakespeare for its awesome complexity, when it’s mostly about as complex for its time as the average Firefly episode.
But it’s greater for that reason, not in spite of it.
I could go on, but don’t put much credence in things like this as it totally ignores the cross-cultural differences, obsolete and archaic words, not to mention that the complexity tests themselves are not in any way a good method of assessment for information actually conveyed.