OK - I was trying to avoid the too-much tech - but it's all about the timing - and perhaps a little of what we call multipath.
GPS timing accuracy: First and foremost - GPS satellites are all sending the same information - they're aware of each other - and the details of their timing accuracy are classified (GPS was originally built for the Department of Defense). All I'm at liberty to say is that each satellite is in as perfect as technically possible synchronization with every other satellite - they all agree on the exact same time - and the time is insanely accurate by any measure. (I worked on a piece of GPS in an earlier part of my career, so this isn't re-hased info I'm tossing out.)
GPS satellite flight: The satellites are stationed in half-geosynchronous orbit. Unlike a communications satellite or satellite TV that appear stationary in the sky - each GPS sat orbits mother earth twice a day. There are a few reasons for this - but only one I'll toss out here: while you theoretically need three sources for triangulation, with GPS you'll get multiples and your receiver's intelligence can choose for best signal quality of those available in the sky - because of atmospheric effects and maybe a few other factors, not all signals from all visible satellites will arrive to you equal in quality. Remember, even though we're talking digital signals - we care about time and we care about time of arrival. The better skyview you have, the better your accuracy - the signal is hitting your receiver directly.
Tower timing: close enough to coordinate with other towers and lock-on to talk/listen to phones - that's it. No more.
Tower signal: can be affected by any number of terrestrial or man-made effects - interference, you name it.
Tower accuracy: for location, signaling, timing - good enough for phone calls, not deploying anywhere near the accuracy of satellites. Calibration performed as needed/scheduled, not _closely monitored_ in real-time by the Department of Defense.
Why does your tower accuracy vary by time of day and by phone? Terrestrial, calibration, or man-made effects are in play.
Call your wife's cell while you stand right next to her - listen to the delay. Do the same to various landlines. Lots of variance there due to telecommunications switching - and the nature of terrestrial-based signals.
If you gave the job of tower location to the Department of Defense, they'd solve every single aspect of this task - and they'd put the towers in orbit.
If you consider that a radio wave in the atmosphere is kinda traveling near the speed of light - 186,000 miles/second - and then consider that you're ONLY bouncing around from a 1/4 mile to a few miles in location accuracy, you'll appreciate just how very close the timing accuracy of the cell towers already is.
A four mile error can be roughly expressed as being 22 millionths of a second off. Not bad when you think of it that way.
Your phone's ok, the cell network is ok, if you want location accuracy and stability - you simply want GPS.
Does this help any?
If interested, see also -
http://en.wikipedia.org/wiki/Global_Positioning_System#Basic_concept_of_GPS
