This hypothesis has been studied here as a possible explanation of selenogenesis (formation of the Moon). (Forming the Moon from terrestrial silicate-rich material. R.J. de Meijer, V.F. Anisichkin, W. van Westrenen. 2013.)
The only place cited as suitable for spontaneous criticality is the Core-Mantle Boundary (CMB).
The calculations above show that without additional concentration factors, $\mathrm{U}$, $\mathrm{Th}$ and $\mathrm{Pu}$ concentrations in the CMB are insufficient to reach criticality. Additional concentrations can be achieved by a combination of two processes: growth of the relative concentration of the fissile materials by a transient pressure wave, induced by an impact at the Earth’s surface (Anisichkin, 1997; Voronin and Anisichkin, 2001), and/or the development of compositional heterogeneities (de Meijer and Van Westrenen, 2008).
As shown by Voronin (2011) for fissile material at the CMB, an impact of a 100km- diameter asteroid can create a transient pressure increase of several TPa at the CMB, sufficient to concentrate fissile material from a subcritical to a supercritical condition followed by a nuclear explosion. Regarding the development of compositional heterogeneities, it should be noted that small-scale heterogeneities exist in the core-mantle boundary region even today (e.g., van der Hilst et al., 2007): volumes exhibiting both higher-than-average and lower-than-average wave propagation speeds, with diameters as small as 30 km, are now resolvable. Some studies suggest that the bottom of the mantle is partially molten today, forming a so-called ‘basal magma ocean’ (e.g. Williams and Garnero, 1996; Labrosse et al., 2007; Lee et al., 2010).
For another take on natural nuclear explosions, see this paper (Evidence for a large, natural, paleo-nuclear reactor on Mars. J. E. Brandenburg, Orbital Technologies Corporation. LPI 2011 Proceedings.), citing an anomaly in Mare Acidalium on Mars.
It should be noted that nuclear explosions need both critical mass and very low times of reactivity insertion. A gun-type bomb would be the best analogy for low-yield natural explosions on the surface/in the crust, and it is very unlikely to have a naturally occurring acceleration hurl a uranium-rich boulder against a uranium-rich wall.
To detect natural explosions on the surface, one would need to look for fused silica and anomalous $\mathrm{Xe}$ and $\mathrm{Kr}$ isotope ratios at the margins of our instruments' sensitivities.
