First, 37signals has a large (for our field) in-house customer support team, so the mean fully loaded head count cost of a 37s employee is lower than the number you're thinking.
Second, I think you are underestimating just how profitable 37signals probably is. They have scaled a few tight products with high recurring revenue to a very large number of customers with a tiny team. I've worked at companies with hundreds of employees and knew their revenue numbers and the model 37signals is executing puts those numbers well within reach.
As an aside to this discussion: If they are actually that profitable it's a good idea to fly below the radar and resist making statements about how wildly profitable they are. Doing that might entice competitors to enter the same market and regardless of whether they succeed or not that could kill the goose.
Sometimes of course things are out in the open and it can't be avoided (lines at the restaurant or with public companies, or Oracle posting their 24x7 software support rates where a guy out of his house can do the math). Sometimes you want the publicity for one reason or another. But despite the openness of the internet there are still things that you want to keep to yourself lest everybody and their uncle decides to move into your neighborhood.
Second, I think you are underestimating just how profitable 37signals probably is. They have scaled a few tight products with high recurring revenue to a very large number of customers with a tiny team. I've worked at companies with hundreds of employees and knew their revenue numbers and the model 37signals is executing puts those numbers well within reach.