The gain in efficiency by running at a higher frequency is only that you can use a smaller inductance, which should mean a lower resistance. But you increase loss's, as you have increased switching loss and increased drive loss. In this application, i need to keep loss's to an absolute minimum as thermal issues are at play, so, i am reducing both the resistance of the inductors ( by using physically big ones ) and running at relatively low frequency. The reduction from say 300hz to 70khz, is a factor of 4.5, and the loss's in the fets, reduce significantly. I'm even running two FETS in parallel, to reduce the effective resistance ( and although i increase switching losses the overall effect is a net reduction in power ). Careful selection of the Fets, is also important.
No, it's a knock-on effect:
In an ideal Steinmetz core loss model, loss is proportional to both Bmax^2 and F^2, so increasing frequency while holding L and V constant reduces ripple current but does not reduce core losses.
In a generalized Steinmetz core loss model, the exponents are variable. Real materials have different responses, but for almost all of them, it increases more sharply with Bmax than with F, so that a higher frequency (still keeping voltage and inductance constant) leads to lower core losses.
If you have occasion to play with just pumping square waves across a ferrite core transformer, or any other kind of inductive component, this is pretty easy to see. I've seen ferrites with exponents something like 3 and 2 respectively, so the savings with F (V constant) goes something like f^(2/3). Adjusting the frequency on such a tester, you can really feel the increased core losses at low frequency, especially near saturation.
The reduced losses at higher frequency mean you can get away with lower inductance (higher Bmax) for the same losses, with a smaller core and winding. (Of course, the smaller core will reach a higher temperature rise.)
Copper losses are usually a significant part of the total, even if not indicated by traditional models (after all.. trying to reach any useful analytical conclusion about coil losses is difficult, and rarely generally applicable). The Q factor of a winding (in and of itself, not counting the core and losses) generally goes as sqrt(F), so that coil losses are lower at a much higher frequency, too.
Making any sort of analytically reasoned argument is really pretty crappy though. You have to check out parts yourself to know for sure. I'd suggest trying out some numbers on Coilcraft's website, because they have a loss calculator that operates directly on their database search, very swanky. The numbers from some of their newer powder composite type coils are pretty slick, and very compact besides.
Tim