As Mihai says, caustics on speculars, among other difficult light propagation situations, is in research and have never been forgotten. Solving it unbiasedly, is an ill conditionated problem, it roughly means that is in its mathematically nature that is hard to solve, and still is an open problem, as in all these years of modern computing no one has found a solution yet.
Yes, there are tons of papers showing some difficult caustics trough glass problems nicely solved, but all of them became very inneficient as soon as the scene shows few or none of these hard features, and they present difficulties as soon as the scene shows simultaneously different lighting complexities. Some programs are able to solve some very specific subset of the problem, under very known conditions ( such as for example not allowing a material to have roughness, having very low ior without dispersion, etc ), and others try with very well known algorithms from those very well known papers, but these optimizations usually introduce penalties in the rendering time when the feature is not present in the scene, and in general introduce new problems in the benefit to solve others.
We can't wait to out maxwell 2 any undefined amount of time until this concrete problem is solved ( I wish we can
). In the meanwhile we choose to invest our time in researching ways to speed up the render in those everyday common situations that does not show the previously mentioned issue. This way we improved lot of useful situations: from materials made of multiple complicate layers, to hdr lighting, direct lighting, and so on. Depending on the particularity of the scene, the optimizations can range from a modest winning, to some magnitude orders in the case of difficult HDR lighting. A particularly hard optimization to achieve was the one with materials, now in multiple bsdf materials, like plastics, the noise reduction is several times faster then in 1.7.
To optimize mathematical algorithms and code is hard, so to beat Maxwell 1.7, which has strongly optimized code, has not been easy. Is similar to the situation in which the best elite athlete need years to win hundredth's of second, and each time being harder to beat.
Said so, and sorry for this a bit long post, is almost a personal challenge to solve the caustics through dielctrics and similar problems, and be sure we are working on it, among much more other areas, of course.