Technically no, this will not effect the transmission. In fact, because of the difference in twist rate between the pairs, there is a substantial length difference between the pairs. The real question is what are you needing to accomplish that the pairs are different lengths?
#1 You can have 45ns delay skew in a 100m piece of a CAT5E cable. This is the difference between the times the signal reaches the other end on the fastest (shortest) pair and the slowest pair. So, that's 4.5E-08 seconds
#2 The signal travels at the speed of light times NVP (nominal velocity of propagation), which is about 0.67 for CAT5E. Therefore the speed of the signal in a CAT5E cable is 2.01E+08 m/s
#3 Multiplying it all up, we have 4.5E-08 x 2.01E+08 equals 9.045 meters allowed difference between the lengths of the shortest and the longest pairs. This is NOT the cable jacket length; the cable jacket length is less than that because the pairs are twisted. For simplicity we'll make the cable jacket length equal 5 meters.
How much of a difference a 3 inch (0.0762 meter) piece can make if you are already allowed to have 5 meters difference? Not very much.
HOWEVER, by saying that you are making pairs 3 inches different in length, you are saying that you have to take the outer jacket of the cable off for at least 3 inches (plus whatever you need for your terminations). This is a big no-no in cabling installation because it creates problems with pairs mixing with pairs of other cables, over-bending, un-twisting and all that bad stuff that can happen to a pair when it's not protected by an outer jacket. The signal might get killed by excessive NEXT or FEXT or any other parameter you pick before you even get to the point of calculating the difference in delays.
So, with that said, if your application allows, don't make them 3 inch different in length. If it does not, make sure the pairs' twists are undisturbed, and different cables' pairs don't mix and you'll be OK.
Thanks for your reply. What I'm trying to achieve is to make a patch panel for gigabit ethernet, where I need to use board-to-board connector on a PCB, and route the signals from a RJ45 jack to another jack, using twisted-pair ribbon cable and the cable is connected in PCB.
So I'm wondering is that critical to have absolute the same length from pair 1 to 4. 1000BaseT does have buffering and error correction, so if I'm going to use it as normal Ethernet data transfer I would think that it does not quite matter that much.
It seems very weird anyone in the world would ever do this, but due to a lot of constrains in our system, we'll like to try this out.