Just for anecdotal evidence... on a recent hard ride 102 miles with 12,000' of climbing, total time 5hr40min, Polar's "own Cal" estimated I burned 2530 Cals. By power data, I had done 5271KJ of work... which by Rics eff estimate puts me somewhere between 4400 and 5500 Cals burned. Looking over lots of exercise files, Polar consistently underestimated by a factor of 2, and this is with my correct height/weight/activity-level/age, etc.
I was trying to lose a couple lbs safely, counting calories, but losing weight way faster than I should have been (I had a spreadsheet to sum total calories, theoretical weight based on surplus/deficit, compared to actual weight) and ended up binging gaining back, being on a yo-yo diet. Initially I figured "well you are as lean as you can get without incredible determination" but... now that I do the same thing with power data, my theoretical and actual weight are consistent (with normal variance of actual weight due to hydration, etc.) and no binge fests brought on by 10,000 Cal weekly deficits.
If you are 5-10 lbs over weight it probably doesn't matter that much, but if you are already 7% BF, and trying to lose and maintain the last couple lbs, you need your caloric intake/expenditure to be as accurate as possible, and the bottom line is the Polar estimate is way low for my cycling efforts.