If you look at the comparison map you'll notice that many old style multipolygons seem to be rendered okay even in the map that ignores old style multipolygons. This has something to do with rendering order.
Look at the following examples. There are two multipolygons. One (gray) area with a hole in the middle and a (green) area that fits inside this hole. In the first two cases (A and A') the first area is correctly modelled with a hole, in the second two cases (B and B') the first area is not correctly modelled, there is no hole there.
Correctly tagged multipolygons (A and A') will always show up on the map correctly. It doesn't matter if the gray area is rendered first (A) or if the green area is rendered first (A'), you will always get the correct result.
In the other case it depends on whether the gray area is rendered first and then the green on top of it (B, correct result) or the other way around (B', grey area hides green area).
The main OSM style orders multipolygons by area before drawing them. Multipolygons with larger area are drawn first. So wrong tagging will often not be visible on the map.
Why is this important for the old style vs. new style tagging?
A multipolygon relation with only a single outer way tagged in the old style (ie. tags on the way) will often appear to be correct on the map. The relation doesn't have any tags, so it will not be rendered at all. The outer ring has the tags and, because it is closed, will be detected as a multipolygon and is rendered. Then the inner ring will be detected as a multipolygon in its own right and rendered, too. Because the outer ring is larger than the inner ring, it will be rendered first. So everything appears to be okay, even though the data is not.
All this only works because of the defined rendering order. And we should not rely on the ordering of the multipolygons before rendering, because not every map will do it this way. And the ordering takes time and effort, so we should not require it.