Well, intuitively I think an annular radiator should be quite good, as it doesn't increase the frontal area of the aircraft!
While I'm no aerodynamicist, I've spent some time thinking about different radiator layouts over the years in search of some general takeaway. And beyond A) The Mustang radiator is really, really good, and B) It's complicated, and details matter, I have largely failed to come up with some simple Grand Unified Theory of radiator layout.
Some general principles can of course be derived from basic fluid dynamics which sort-of leads to the Meredith effect. Namely you want a diffusing plenum to reduce the velocity of the air so that the drag through the radiator matrix is reduced. And the opening angle of the plenum shouldn't be too high lest turbulent flow is formed. And similarly after the radiator matrix you want a constricting nozzle, ideally producing some thrust. And then you have considerations like you want to minimize the surface area of the ducting to reduce skin friction, which implies you want as close to a circular cross section as possible. And the size of the radiator also depends on the speed regime you're targeting. For a higher speed design you probably want a big radiator with equally big ducting to get the velocity of the air stream down, whereas a lower speed design can make do with a smaller radiator + ducting and save a bit of weight and skin drag. And finally there are of course non-drag related considerations like having the radiator and engine close by makes it harder to score a critical hit, and potentially allows power egg style approaches, as well as reducing the length of coolant piping between the radiator and engine.
So in light of the above, what can we say about a few radiator designs:
- The Mustang belly radiator is pretty much the gold standard. But it doesn't follow that any belly radiator in general is a good design. E.g. the belly radiator for the Tornado prototype was so bad they switched to a lower drag chin radiator for the production Typhoon and Tempest V, hardly paragons of low drag radiator designs? Again, details matter, and just by eyeballing it the Tornado belly radiator looks more like an enlarged Hurricane radiator than something that would remind one of the Mustang.
- Leading edge radiators were evidently considered a success on the Mosquito. But then they apparently also tested a power egg style installation (basically to the Lancaster engine nacelles with chin radiators), which resulted in a pretty small performance penalty.
- For annular radiators, the Napier tests with the Tempest prototype were quite favorable, with the annular having less drag than LE radiators as well as the Tempest V style chin radiator. But if one compares an annular design with, say, the Mustang radiator it seems there's not much room for a big diffusor plenum and nozzle, as well as the exhaust not being directed straight back but rather must be directed at an angle, reducing the thrust effect somewhat. And somewhat confusingly for the post-war Tempest follow-up designs (that never left the paper stage), Hawker studied both LE and belly radiators (P. 1027 and 1030), but not annular ones. So what gives, wasn't the annular design as good as the Napier reports suggested?
- Wing radiators not on the leading edge like the Bf 109 or Spitfire. Maybe not as good as (well implemented) above approaches?
I think the "holy grail" might be something that would combine a Mustang-style diffuser + nozzle ducting along with a compact package that could be used for a power egg style installation, suitable for nacelles in multi-engine aircraft as well as single-engine planes. Say, have the air intake just under the spinner, the diffuser duct under the engine, and then the radiator matrix would be behind the engine. As an additional bonus, combine the radiator air intake duct with the supercharger air intake, thus the supercharger also gains a bit of pressure due to the diffusing duct, and by having the intake at the very front of the aircraft the chance of ingesting all kinds of crap is also reduced. Something like Figure 11 in
https://www.arpnjournals.com/jeas/research_papers/rp_2015/jeas_0715_2249.pdf .