Depending upon satellite geometry and the number of satellites 'visible' at the time of observation, an un-augmented single GPS receiver can only deliver accuracies with around 10-30 meters. GPS receivers are also considerably weaker in elevation (Z) computation than they are in X and Y. Now, you probably will get much better accuracy, but statistically speaking as long as it's within that 10-30 meter circle, those computed coordinates are about as accurate as you'll get for a single observation. Augmenting these computed GPS coordinates with other stuff like cell towers or known locations of nearby wifi stations allows our cell phones to get within a few feet (~1-2 meters) most of the time, but without going to a survey grade GPS unit with differential correction, you'll have difficulty getting repeatable accuracy better than 1-2 meters. A survey grade GPS with differential correction can get you into millimeter accuracy range, but nobody would want to ride with a device that bulky, cumbersome, inconvenient, and freakishly expensive.
Since these apps/devices that track our rides are computing distances based upon a series of individual observations, each with a potential error of several feet in any direction (including up/down), you can see why you'll never get perfectly matching distances for a nearly identical route measured multiple times.
My question is, do you actually need accuracy less than a foot per observation? Consider that your phone/device is taking faint signals from a bunch of super-accurate clocks spinning around the globe at 14,000km/h and somehow crunching that down to an X,Y,Z location. I'd say that getting within a few feet without any measuring devices at all is pretty darned impressive. Probably best to just enjoy the ride and not worry about a few feet of difference.