In chapter 10.3. Selection in worst - case linear time ($k$th minimum) from Introduction to Algorithms by Cormen, Leiserson and Rivest, the time complexity expected for step 5 of the algorithm presented in this chapter is said to take $T(n)=7n/10 + 6$ at worst.
From my point of view the time complexity is the one specified above because the number of elements larger than the median of all medians is at least $3n/10-6$, thus the remaining search space is $7n/10 + 6$, where $n$ is the size of input data.
How off am I?