MPT should be called Medieval Portfolio Theory, it is a theory from 50 years ago with huge theoretical flaws (mean-variance utility, use of Pearson's correlation that is not coherent, based on historical data). Come on, it is an error maximizer. The least one could do is Michoud resampling, but it is patented. Or a bayesian Black-Litterman would be more appropriate.
So to answer the question: no, anyone reasonable in the world of asset management won't use MPT. Of course one should care about higher moment, asymmetries, using conditional measures and robust/coherent risk measures. It complicates hugely the formulas though, where a closed form formula or "beautiful solution" may not exist, and we should rely on quantitative modelisation that works.