A lot of misinterpretation of theory is evident here.
Current thinking is that exercise behaviour is regulated through knowledge of the endpoint of exercise (the process is termed telioanticipation). This knowledge is combined with all sorts of feedback on the current status of the physiology (glycogen levels, oxygen saturation, pH, temperature etc) to determine an appropriate muscular workrate that will not result in some kind of catastrophic physiological failure before the end of the event. Workrate itself is basically determined by the muscle mass the brain is willing to recruit. So, you may start a race at what seems like a sensible pace, but some distance into it afferent feedback to the brain says that e.g. muscle glycogen stores are dropping (or you are getting too hot, pH is dropiing etc etc). The brain does a quick calculation and realises that given the rate of use of glycogen and the distance still to cover, then there is no way that the body can maintain the current running speed untill the finish line without damaging itself in some way. In order to prevent this happening, less muscle mass is recruited, thereby reducing the rate of glycogen use, but also reducing the running speed.
Now - obviously when running you do not think to yourself "hmm, glycogens getting a bit low, I'd better slow down". This feedback is proposed to be processed at the subconscious level and you simply experience 'fatigue' in the sense that maintaining your current pace becomes increasing more uncomfortable. A lot of work has looked at the role of the Rating of Perceived Exertion (RPE) in regulating behaviour and what is consistently found is a linear increase throughout exercise culminating in the highest values at the very end. The RPE athletes rate is 'conscious', but this conscious RPE is suggested to be compared to a subconscious 'template' that serves the purpose of preventing self harm. In a race therefore you may start quickly to keep up with rivals, but if this pace is too fast for your physiology to maintain then your (conscious) RPE will quickly increase rapidly above the levels your brain is happy to allow. You therefore experience discomfort untill you slow down to a speed your subconscious brain is happier with (Your brain is 'happier' because it knows that there is no risk of self harm if you maintain current speed to the finishline).
Imagine you are a 30 flat 10k runner. If you were asked to run 1k on the track in exactly 3 minutes I'm sure that if I asked you afterwards to report your RPE it would be low. If I asked your RPE after the first 1K of a 10k race also run in 3 minutes then I would expect it to be a little higher, and if I asked you after you had covered the first 1k of a marathon race in 3 minutes I would expect it to be almost maximal. Why is this? Clearly the actual work you have performed is identical (1k in 3 minutes). The reason is because you have interpreted the feedback from your physiology in the context of what you know is still to come. So if you covered the first 1k of a marathon in 3 minutes your brain is screaming at you to slow down because it knows it cannot maintain this workload for another 41K.
Therefore the simple answer to the original posters question is - because your brain calculates that you cannot complete the marathon at the same speed without risk of some kind of physiological catastrophe, and does everything it can to make any attempts to do so as painful as possible!