You're right -- the numbers don't matter that much. It's always good to see candidates have solid general knowledge when computing their assumptions, but it's a bit odd if they happen to know how many golf balls are sold in France every year off the top of their head.
My suggestion is to start from the basics that everyone does (or should) know -- population size, people per household, age distribution trends, etc. From those, you should be able to make 'laddering' assumptions that essentially convert the units from things you know (number of people) to things you want to know (how many golfballs sold in France).
So, population of France is about 70m. Assuming that 10% of the population plays golf at least once a year, that means 7m people play 18 holes. Lets say that, on average, there are 1.5 games per person per year. So, now we're at 27 holes per person per year, times 7m. If a ball gets lost every 9th hole, that means each person goes through 3 balls per year. 3 times 7m is 21m, which suggests 21m golf balls are sold every year in France.
That sounds high to me, so at that point I'd ask the interviewer if any of my assumptions are terribly off.
Again, the point isn't to know a bunch of truly random trivia, but rather to show a good broad knowledge base and facility with both computing and communicating simple math. After all, in a real world setting the thing we're measuring for is "will this person be a good calculator and collaborator on market size estimations? how can I tell without that person having a computer in front of them today?"
The ability to go through an example like this confidently, quickly, and with clear communication far outweighs whatever brownie points you might score for getting close to the right answer. So focus on that.
Hope this helps!