Boden & Bagnall (pdf) looked at this question of reflection, in their paper Bio-Mimetic Subwavelength Surfaces for Near-Zero Reflection Sunrise to Sunset.
Their estimate of the proportion of photons reflected, from sunrise to sunset, in a fixed PV system is 20%, rather than your estimate of one-third.
The proposed etalon is not an efficient or effective means of capturing the energy in reflected photons.
However, there are not-dissimilar ideas being researched at the moment, including a lot of work in plasmonics.
Boden & Bagnall propose a moth-eye style coating. Their simulations predict an increase of 12% in energy output, compared to single-layer anti-reflective coatings:
(source: paper as linked above)
However, PV panels are now largely a commodity product that compete on price (except for some specialist applications, e.g. satellites) - so it's generally cheaper to buy more panels, than to pay for fancy coatings on fewer panels.