I would not consider this a bug as the plugin runs excatly the way it is intended to. The very concept of Iso-areas implies that whole surfaces and not just networks (lines) are covered, thus even areas that do not line on the network. This in itself, conceptually, does not represent reality 100% as we all know from our everyday experience. Technically, if you run a network analysis, you can only reach points lying on the network lines, everything else, beside the network, is just interpolated. So you should rather set a clear critearia what means "reachable" in a Iso-area network analysis: how far away from the network (lines) is still "reachable"?
Consider a real street network in a city: you can't reach all areas that lie outside the street itself: private courtyards, a closed compound of a factory etc. The same is true for a lake, river, a cliff etc.
So rather than consider this a bug or malfunction, it's more convenient to critically reflect the very concept of Iso areas. They are a model that does not necessarily correspond to an actual reality - in some cases yes:parks are areas perfectly reachable, in other cases no: the compound of a prison is not (normally) accessible, even if it is surrounded by a network of streets that is reachable.
The plugin has no way how to differentiate areas reachable and areas not reachable. If you want such an output, you should create another polygon layer that covers areas reachable and areas not reachable (or a raster layer to differentiate different steps of "reachability", like from 1-perfectly reachable to 0 - not reachable at all). Than you can overlay this with the output of your Iso-area analysis to get a more suitable result - like clipping the Iso-area with a polygon representing the water surface.