You could calculate the average performance difference caused by the difference in file size as the time it would take to fetch one more TCP/IP package times the probability that it would happen because of that change (i.e. package size divived by the number of characters added).
You might get something like 10 ms * 1/1000
, which would give you 10 µs
.
That is such a small performance difference that you have to make the same thing a huge number of times before it's noticable.
So, you should use the one that is clearer and easier to maintain.
Personally I would go with the first option. I find it easier to see what's applied to the element if there is a single path to follow from the class names to the rules, rather than having a class scattered across several rules.