90% can be below average. It depends on how you define 'average.' If you're using the median, then 'below average' can only be 50%. But if you use the mean (or its variants), then you can have 90% of a population be 'below average' if the population is very skewed.

Also, he was talking about 90% of coders he interviews. Of course they'd be below average - if they were above average, someone would hire them and they'd leave the job poll quickly. Such is the pitfalls of sampling a non-random population...

I added the "seriously though" to clarify the obvious joke. Sorry you took it seriously, though.