76%. I tracked down the story. In the below book in chapter five "The Genius at the Royal Mint". They simulated ten seasons worth of games 100,000 times, basically doing the above method and looking for how often an 'incident' of that size happened. Which apparently was a surprising 23% of the time. AKA over the entire system, heads tails evens out to 50/50, but clumps of runs can happen in such 'high' percentages. I guess it's sort of intuitive... if you sum up the possible outcomes and 50/50 prevails, then if you go on a 75%-ish win streak a quarter of the time, you'll go on a 75% losing streak another quarter of the time.

Thinking about it further, if a run like Chelsea's happened, it'd be counterbalanced by another team or teams going on a run of 'opposite' character, AKA drawing stronger opponents (since Chelsea removed a weaker one each time)

https://www.goodreads.com/book/show/39328092-aiq