The GPS chips in phones have gotten really good at fixing a static position-- most new phones can get within a few meters. And their refresh or polling rate is 1hz, just like pretty much every consumer GPS device there is (there are some devices out there with 5Hz up to 20Hz rates) so they have the capability to be every bit as accurate as a dedicated GPS device. But the power requirements of GPS is one of the highest drains in a phone-- just behind the screen-- so the programming (and sometimes the hardware itself) quite literally cuts corners in an effort to balance GPS use and battery life. Depending on conditions, reported position can fall to the minimum accuracy of the chip-- 300m. This is why GPS is not great at determining speed. Even on a bike at 30mph, the GPS refresh is occurring once every 44 feet. So if a segment is a scant 0.2 miles long and the phone user gets the start/finish positioning "bonus," he will cover 8% (or more) "less distance" on that effort. The start/finish of a segment might happen to fall in the middle of a refresh for the GPS chip, so the rider is a full second (or more, based on the accuracy of the position correction) into the segment before the phone records that he's there. This is (in my estimation) why I see segment efforts with an average speed higher than the max speed.

Apps like Strava can't override the phone's OS instructions for the GPS, so you get inaccuracy. I've seen some files with insane zig-zags that cross over entire neighborhoods, and pick up segment PRs / KOMs on parallel streets to what the ride was actually on. This isn't as good, but it shows the less precise position correction of a phone GPS compared to a dedicated device. Strava Android App on the top, my 520 on the bottom: