Ljungqvist and Sargent (2004). Recursive macroeconomic theory 2n ed. (ch. 15) present and review the issue. In the Concluding Remarks section, they mention two environments, where the "zero-optimal-capital tax rate" does not hold:
Aiyagari(1995) presents a model with heterogeneous agents, incomplete insurance markets and borrowing constraints (i.e. a "Bewley" model): the optimal capital tax rate is positive, even in the long run. Insight: incomplete markets lead to excessive precautionary saving, leading to over-accumulation of capital. The positive capital tax rate offsets that.
Golosov, Kocherlakota and Tsyvinski (2003) introduce private information (the government cannot observe the hidden skill levels of different households). The tax system becomes an optimal dynamic incentive mechanism, leading to positive optimal capital tax rate.
Note that the result depends also on the feasibility of government commitment. If a commitment mechanism is not in place, it will be optimal for the government to "renege on its promise and levy a confiscatory tax on capital", as L & S write.