The primary problem you're encountering is that the Large Scale International Boundary table contains several very complex polygons and those complex polygons take a long time to reproject (e.g.: the Canada polygon contains 193k vertices).  That's compounded by the fact that the reprojection is repeated for every tile that's processed, so even though the reduceRegion(s) doesn't fail, it can take a (very) long time to complete due to the extra overhead added to every tile.

For extremely long aggregations like this, the mapped reduceRegion is better because each feature that's processed gets cached, so if you end up having to restart, the intermediate values aren't lost.