I wouldn't use offensive humorous test data, but using obviously fake humorous test data makes it easier to detect and mitigate those kinds of situations. If for some reason I noticed that all of my production data came from an obscure browser that called itself "The Thrilla In Mozilla", that would be a lot more noticeable and easier to correct than if my test user-agent string was just Chrome or whatever.