As Howard Marks puts it, for most investors the only real measure of risk is the permanent loss of capital. That occurs for two reasons: investing in something that can become worthless, or an emotional response at an inopportune time, preventing mean reversion to occur for a given allocation. MPT is not fatally flawed, but mischaracterized. It is not designed to prevent or eliminate losses! Even the comment that the variance/covariance matrix is unstable may be a reference to 2008, when correlations theoretically went to 1 (for a brief period of time). However an investor who owned stocks AND bonds lost less money than an investor who only owned stocks, proving that MPT works. An investor's appropriate risk measure is substantially dependent upon their time horizon. And even then, not necessarily their investing time horizon, but their emotional time horizon. An investor who has the ability to monitor their portfolio less frequently will find they can tolerate a lot more risk.

The fatal flaw is in characterizing MPT as a volatility elimination tool. Earning return requires a willingness to bear risk. MPT provides a defined methodology to determine what level of risk / return tradeoff an investor can tolerate.