I can't reproduce the scenario you described.

Based on your sample data the code runs as expected (I'm purposely not saying it runs "fine" 😉 ).

If you already tried several filters (including LOCAL STATIC FORWARD_ONLY) and you still don't want to rewrite it as a set-based solution, you could try to narrow-down the root-cause by assigning a NEWID() to the outer sproc and include that column into the lpn table. Therewith you'd be able to verify that the sproc itself is going crazy or if it's called multiple times.

My question reagrding the process in general:

Why isn't there any WHERE condition to ensure that only rows without an already assigned lpn value will be selected?

What I've seen when runnig your code: If you call it twice, then it will update all rows with country = 'HR' even though those rows already have an lpn value assigned. Is that intentionally?