I'd certainly try tossing them into a single ImageIntegration run. Pick a good 3nm sub as the reference.
The other thing I'd try is doing a stack of 3nm and a stack of 5nm and then combining the stacks using ImageIntegration without rejection (you'll need to supply two copies of each.)
Both those options will favour overall SNR. If you find that the 3nm data gives substantially better contrast then the third thing you might want to try is blending the two stacks with PixelMath giving up some SNR for improved contrast, say an expression like "0.25*5nm_image+0.75*3nm_image". Adjust the 0.25 and 0.75 scaling factors as need.
Cheers,
Rick.