The literature on optimal income taxation kicked off with Diamond, Mirrlees (1971). In their model, people differ only by ability, i.e. the highest earning person is the most productive one. As a consequence, any marginal tax rate on this person will be detrimental to production and hence to social welfare (in this model). Therefore, the optimal top marginal tax rate is zero!
The main innovation was initiated by Emmanuel Saez (Saez (2001) and Piketty, Saez (2012)). The basic idea is that the social planner maximizes a social welfare function (SWF) that captures the fact that individuals are to some degree inequality averse. Plus, they allow the introduction of welfare weights, i.e. a normative decision on whose well-being matters for the society as a whole. The innovation of this literature is the formulation, that allows to calculate an optimal tax rate based on observable (or, rather, estimatable) figures. This is referred to as the sufficient statistic approach. See e.g. Eq.7 in Piketty, Saez (2012) for the optimal top tax rate $\tau$:
$$\tau = \frac{1-g}{1-g+a *e} $$
This depends only on three parameters: $e$ is the extent to which top income earners are able to dodge the tax. In the basic model, this is only labor supply, but later contributions (incuding the one you cite, but also Saez et al. (2012), [yes, Saez again]) acknowledge that there are many other avoidance strategies. $g$ is the social welfare weight of the top-income earners and $a$ is an indicator of the income distribution.
The optimal top marginal tax rates above 70% stems basically from the assumption that $g$ is zero, i.e. society as a whole does not really care about the richest. Another key factor is the magnitude of $e$. If I argue that with a sufficiently high top tax rate, all millionaires will leave the country, I assume a high $e$, which drives down $\tau$. There are also researchers that came up with different values for $a$. One quite well-known criticism is from Mankiw et al. (2009). They back up their arguments however, e.g. on less reliable data, which is why I don't find it very substantiated at least on the top tax rate issue.
For a general discussion on optimal taxation beyond the top tax rate, see also Diamond and Saez (2011).
Summing up, the Saez literature is a great tool to link data and economic theory and has been heavily used. Empirically, estimating the various margins of $e$ is very difficult and has been tried numerous times. The top tax rate of 70% or higher is debatable as it has some normative settings. But overall, this literature has made compelling arguments in favor of progressive taxation and higher taxation of inheritances. It might be my personal bias, but I'm not aware of any fundamental, substantiated criticism to the Saez approach.