The most prototypical evidentials are a class of verbal affixes. They are not commonly tenses or moods of their own (though they can be). As such, they are not structurally incompatible with the irrealis/conditional, and ultimately it really depends on what exactly that mood is being used for by a language.
In Quechua, counterfactuals (If I had hooves, I'd be a horse) use the irrealis marker -man, but cannot (as I understand it anyway) take evidentials, but a statement of probability (She may be at school), which also uses the same marker, definitely can.
Irrealis statements about future events or hypotheticals, on the other, can also reasonably be said to be semantically unsuitable to use evidentials: in I will try to feed her some soup, the "feed her some soup" subclause is not really a statement you can really describe the source of the information (unless you have a dedicated evidential for that case, but natural languages don't). Nonetheless, WALS notes that some languages do:
Because evidentials are used to describe the speaker’s involvement with events, they tend to occur in realis contexts, especially in past tense situations. Nevertheless, evidentials do occur in what can be described as irrealis situations. Example (9) from Barasano (Eastern Tucanoan; Colombia; Jones and Jones 1991: 116) shows an interrogative evidential, and (10), from Tsova-Tush (Nakh-Daghestanian; Georgia; Holisky and Gagua 1994: 180) shows an evidential with a future event.