What are the input/ output voltages please?
Kool Mu -26 is about 0.1 watt per cc at 25kHz, 100mT peak AC excitation, 450mW per cc at 0.2T peak excitation, which is getting up there, essentially no core losses for DC or 50/60Hz operation. So ideally keep the L as high as you can to keep the ripple current (and hence ripple B) down to 0.02 to 0.05T peak AC excitation according to the loss graphs..
It is a slightly iterative method to design with these cores, you need to pick your lowest acceptable L for a given peak I (including ripple current) and work from the tables to find a (large in your case) core, the A.turns per mag path length, and work back to find the turns, work out the resistance of the turns that will fit and see if the R is too high.
If too high go to a larger core... usually operate there cores at 50% saturation at the worst case Ipk.
Then you can calc the ac core losses...
You may need to parallel inductors if there isn't a large enough core to keep the wire losses down to reasonable...
Starting with the largest Kool-mu toroid, 165mm OD (2.2kg) , the -26 has a nominal Al = 78nH/t^2, giving 113T for 1mH
the AT are therefore 2260 @ 20A, giving an effective AL at 20A, of 70nH/t^2 i.e. 893uH at 20A thru.
The ID = 101mm, for a single full winding you can use up to 2.8mm dia wire,
putting into my winding program, if we use 2.62 dia wire (better fit), the MLT = 134mm, giving 19 watts at 20 amp at 100 degC for the 113 turns. Lil bit hard to wind, but can be done...Still pretty high but maybe acceptable if there is air flow.
OK, so core losses at 25kHz, you say ripple is 2.5A pk on 20A at 25kHz, 1mH, from the graphs, the losses will be (B @ 2260 AT/m (41.2cm) = 0.18T so the peak flux ripple is 0.0225T, gives Pcore = 6mW / cc x 407cc = 2.5W approx. If go to more layers on the toroid can use bigger wire to get the Pcu down. (Or more strands of a lighter wire - easier to wind..)
Two cores glued in parallel gives Al = 156 nH/t^2, and using 100 turns gives 1.52mH (@20A), and a corresponding drop in ripple flux and cores losses, and allows wire to go to 3mm dia, (Pcu = 19.8W @ 20A).
Core losses will be a little less - say 2W approx. Just more surface area to get rid of the Cu losses...