As EEs we apply Kirchhoff's laws to solve circuits. Which do not apply in every case. Really what works in every case are Maxwell equations which are a nightmare to solve even in simple circuits. KVL and KCL are simplifications of MEs when certain conditions are met: total electric charge is constant, no linkage of magnetic flux, low frequency compared to wire length (no radiated energy). We must comply with the lumped element model.
The "ideal" components that we use in our circuit analysis are defined in the LEM.
When we try to solve circuits with zero resistance, ideal capacitors and ideal voltage sources. Strange things occurs. The capacitor gets charged
by infinite current in zero time. Yet the product of an infinite current by a zero time gives a finite amount (the charge). When two capacitors are connected together, a finite charge is transfered in zero time.
When we introduce finite resistance things get more manageable, equations dont get weird and we are much more near to what we see in practice.
My point is that we need to be very carefull when we use a model and push it to mathematical limits of zero or infinity. Otherwise we may violate the conditions that made valid that model.
In the context of Kirchoff laws and ideal circuit elements (RLC), the only way that energy is lost is by heat in a resistor. There is no EM waves, sparks or any other means. To take into account this effects we need to apply Maxwell equations or add parasitic components to our analysis.
For example, the analysis of an antenna or a transmision line.
In the case of the two capacitors I like more the following explanations which do not involve radiation. (The 2nd link explains why not)
http://arxiv.org/pdf/0910.5279.pdfhttp://arxiv.org/pdf/1309.5034.pdfThis thread is very interesting. Hope that I reached two cents