> Is there some fractional-wavelength propagation?

"Propagation" is probably not the right word, but fractional wavelengths can "leak" some amount of field.

https://en.wikipedia.org/wiki/Evanescent_field

Notably, that field is powerful enough that microwave doors need to be about an inch thick. The rest of the inch is just empty space and a sheet of plastic to stop you putting your fingers into the evanescent field.

Unfortunately there were a few microwaves sold recently in Europe which forgot that design element, and if you pointed at food while cooking it, you would cook your fingers too.

You're right. The rule of thumb for EMC engineers is to bond joints at no further than wavelength/10 where the wavelength corresponds to the highest frequency that you wish to maintain good shielding effectiveness. Some MIL projects use wavelength/100.

But the gap around the door might only be millimeters wide, but it is 30cm or more long.

It is "both* dimensions that must be a long way below the wavelength to keep microwaves in.

Instead the door seal uses a technique called a quarter wave choke. It relies on reflecting back any microwaves wanting to escape, and by making incoming and reflected waves perfectly cancel, no power is transmitted.

As OP said, even a small amount of soup dripped onto the seal and your microwaves will all start escaping.