and normally it's true
I think this is fairly consistent with how maxima behaves. a-b doesn't instantly return 0; you have to apply some transformations (radcan) to get 0.
Marking as pending.
Well, this is a real bug -- if Maxima can't determine the equivalence of the two radical expressions, it should return unknown instead of false.
(It is a lesser bug that it can't determine the equivalence of the two expressions. But the main problem is the spurious false value.)