I do not see any reason why a small strait cannot be “inside” a large strait.
Slightly off topic: Similar issues have previously been mentioned about bays, fjords, peninsulas, etc.
The issue is that if we replace it by a node or a single line, the importance/size of the strait/bay/fjord/peninsula is much harder to define.
When it is made as a polygon or multipolygon, you can quite easily calculate the (approximate) area of the feature and use the size to determine at which zoom level you should render the name and e.g. how large or dominant the name should be.
For an example, look at how Reitstøen managed to do that for the various parts of Jonsvatnet: mtbmap.no: Jonsvatnet (when you zoom in, names will appear for the smaller bays, straits and peninsulas).
Similarly, for Trondheimsfjorden and its multiple bays of various sizes: mtbmap.no: Trondheimsfjorden (Trondheimsfjorden and Strindfjorden appear at z8, Korsfjorden and Gaulosen appear at z9, but Ilsvika, Korsvika and Leangbukta appear first at z13 since they’re a lot smaller)
These relations do not have to follow the coastline perfectly, so for large features like Sognefjorden it can be a very simplified polygon instead of a very complex multipolygon. That reduces the maintenance issues.
Personally I’m a fan of doing the naming of e.g. woods and parts of woods in a similar way, with simplified multipolygons. Then the rendering engines can determine when, where and how to render the names of each area, such as Nordmarka and Østmarka in this example: mtbmap.no: Oslo
When you start zooming in, you will be able to discover Krokskogen, Lillomarka, Skimarka, etc. in smaller fonts than Nordmarka and Østmarka. This is automated.
I’m aware that some people disagree in how this is tagged.